A Smiling Bear in the Equity Options Market and the Cross ... · A Smiling Bear in the Equity...
Transcript of A Smiling Bear in the Equity Options Market and the Cross ... · A Smiling Bear in the Equity...
A Smiling Bear in the Equity Options Market
and the Cross-section of Stock Returns
Hye Hyun Park∗, Baeho Kim†, and Hyeongsop Shim‡
(This verstion: September 20, 2016)§
ABSTRACT
We propose a measure for the convexity of an option-implied volatility curve,
IV convexity, as a forward-looking measure of risk-neutral tail-risk contribution to
the perceived variance of underlying equity returns. Using equity options data for
individual U.S.-listed stocks during 2000-2013, we find that the average realized
return differential between the lowest and highest IV convexity quintile portfolios
exceeds 1% per month, which is both economically and statistically significant on
a risk-adjusted basis. Our empirical findings indicate the contribution of informed
options trading to price discovery in terms of the realization of tail-risk aversion in
the stock market.
Keywords: Implied volatility; Convexity; Equity options; Stock returns; Predictability
JEL Classification: G12; G13; G14
∗Korea University Business School, Anam-dong, Sungbuk-Gu, Seoul 136-701, Republic of Korea, E-mail: [email protected].†Corresponding Author. Korea University Business School, Anam-dong, Sungbuk-Gu, Seoul 136-701,
Republic of Korea, Phone: +82-2-3290-2626, Fax: +82-2-922-7220, E-mail: [email protected].‡Ulsan National Institute of Science and Technology, UNIST-gil 50, Technology Management Building
701-9, Ulsan, 689-798 Republic of Korea. E-mail: [email protected].§This research was supported by the National Research Foundation of Korea Grant funded by
the Korean Government (2015S1A5A8014515), for which the authors are indebted. The authors aregrateful for helpful discussions and insightful comments to Kyounghun Bae, Ki Beom Binh, BartFrijns, Paul Geertsema, Daejin Kim, Dongcheol Kim, James A. Park, Bumjean Sohn, Sun-Joong Yoon,and participants of the 28th Australasian Finance and Banking Conference, 2015 Derivative MarketsConference, the 11th conference of Asia-Pacific Association of Derivatives. All errors are the authors’responsibility.
1. Introduction
The non-normality of stock returns has been well-documented in literature as a natural
extension of the traditional mean-variance approach to portfolio optimization.1 In general,
a rational investor’s utility is also a function of higher moments, as the investor tends to
have an aversion to negative skewness and high excess kurtosis of her asset return; see
Scott & Horvath (1980), Dittmar (2002), Guidolin & Timmermann (2008), and Kimball
(1990).2 While considerable research has subsequently examined whether the higher-
moment preference is indeed priced in the realized stock returns (e.g., Chung, Johnson &
Schill (2006); Dittmar (2002); Doan, Lin & Zurbruegg (2010); Harvey & Siddique (2000);
Kraus & Litzenberger (1976); and Smith (2007)), the higher-moment pricing effect is
embedded in equity option prices in a forward-looking manner. Specifically, the shape of
an option-implied volatility curve reveals the ex-ante higher-moment implications beyond
the standard mean-variance framework, as the curve expresses the degree of abnormality
in the option-implied distribution of the underlying stock return by inverting the standard
Black & Scholes (1973) option-pricing model.3
Although empirical research abounds in examining the risk-neutral skewness of
stock returns implied by equity option prices, the option-implied excess kurtosis has
received less attention.4 Our study attempts to fill this gap. Exploiting the fact that the
shape of an option-implied volatility curve contains information about ex-ante
higher-moment asset pricing implication, we propose a method to decompose the shape
of option-implied volatility curves into the slope and convexity components. Motivated
by stochastic volatility (SV) model and stochastic-volatility jump-diffusion (SVJ) model
specifications, we conjecture that slope and convexity of the option-implied volatility
curve contain distinct information about future stock return and convey the information
about risk-neutral skewness and the excess kurtosis of the underlying return
distributions, respectively. We confirm that the slope and convexity components (IV
slope and IV convexity hereafter) carry different information from extensive numerical
1The mean-variance approach is consistent with the maximization of expected utility if either (i)the investors’ utility functions are quadratic, or (ii) the assets’ returns are jointly normally distributed.However, a quadratic utility function may be unrealistic for practical purposes, as it exhibits increasingabsolute risk aversion, consistent with investors who reduce the dollar amount invested in risky assets astheir initial wealth increases; see Arrow (1971).
2Decreasing absolute prudence implies the kurtosis aversion according to Kimball (1990).3To better explain the positively-skewed and platokurtic preference of rational investors, prior studies
have attempted to relax the unrealistic normality assumption to capture the negatively-skewed and fat-tailed distribution of option-implied stock returns by extending the standard Black & Scholes (1973)model to stochastic volatility models and jump-diffusion models.
4There are some notable exceptions from this trend; refer to Bali, Hu & Murray (2015) and Chang,Christoffersen & Jacobs (2013) for example.
1
analyses and investigate whether the risk-neutral kurtosis, proxied by IV convexity,
predicts the cross-section of future stock returns, even after the option-implied skewness
effect is controlled.
Using equity options data for both individual U.S. listed stocks and the Standard
& Poor’s 500 (S&P500) index during 2000-2013, we study the ability of IV convexity
to predict the cross-section of future equity returns across quintile portfolios ranked by
the curvature of the option-implied volatility curve. We find a significantly negative
relationship between IV convexity and subsequent stock returns. The average return
differential between the lowest and highest IV convexity quintile portfolios exceeds 1%
per month, both economically and statistically significant on a risk-adjusted basis. The
results are robust across various definitions of the IV convexity measure. Both time series
and cross-sectional tests show that other well-known risk factors and firm characteristics
do not subsume the additional return on the zero-cost portfolio. Moreover, the predictive
power of our proposed IV convexity measure is significant for both the systematic and
idiosyncratic components of IV convexity, and the results are robust even after controlling
for the slope of the option-implied volatility curve and other known predictors based on
stock characteristics.
Where does the predictive power of IV convexity come from? Our finding is
consistent with earlier studies demonstrating the informational leading role of the
options market to the stock market. Extensive research demonstrates that equity option
markets provide informed traders with opportunities to capitalize on their information
advantage thanks to several advantages of option trading relative to stock trading.
Examples include (i) reduced trading costs (Cox & Rubinstein (1985)), (ii) the lack of
restrictions on short selling (Diamond & Verrecchia (1987)) and (iii) greater leverage
effect and built-in downside protection (Manaster & Rendleman (1982)). In this
context, Chakravarty, Gulen & Mayhew (2004) document that the option market’s
contribution to price discovery is approximately 17% on average.5 Although one may
think of the private information as being directional, information on uncertainty can
also motivate informed options trading; see Ni, Pan & Poteshman (2008) for instance.
Then, why is the relationship negative between IV convexity and the realized return of
underlying equity return in the subsequent month? We highlight that the shape of option-
implied volatility curves reveals the implication of risk-neutral higher order moments of
underlying return distributions. Given that excessive tail-risk is certainly unfavorable
to rational equity investors, the fat-tailed risk-neutral distribution incorporates ex-ante
5See also An, Ang, Bali & Cakici (2014), Chowdhry & Nanda (1991), Easley, O’Hara & Srinivas(1998), Bali & Hovakimian (2009), and Cremers & Weinbaum (2010) among others.
2
estimates of market risk premia regarding future rare events. In other words, a heavier tail
in the risk-neutral distribution is associated with a higher required return (or, equivalently,
a larger discount rate), and the informed option trading related to IV convexity predicts
a negative short-term return on the underlying stock returns as a result of price discovery
driven by an asymmetric information transmission from options traders to stock investors.
Simply put, our empirical findings indicate the contribution of informed options trading
to price discovery in terms of the realization of tail-risk aversion in the stock market. We
confirm that the predictive power of IV convexity becomes more pronounced for the firms
with stronger information asymmetry and more severe short-sale constraints, especially
during economic contraction period, and the cross-sectional predictability of IV convexity
disappears as the forecasting horizon increases.
Our study extends recent literature showing an increased interest in inter-market
inefficiency, leading to a proliferation of studies into the potential lead-lag relationship
between options and stock prices. For example, An et al. (2014) find stocks with large
innovations in at-the-money (ATM) call (put) implied volatility positively (negatively)
predict future stock returns. Xing, Zhang & Zhao (2010) propose an option-implied
volatility skew (henceforth IV smirk) measure showing its significant predictability for
the cross-section of future equity returns. Jin, Livnat & Zhang (2012) find that options
traders have superior abilities to process less anticipated information relative to equity
traders by analyzing the slope of option-implied volatility curves. Yan (2011) reports a
negative predictive relationship between the slope of the option implied volatility curve (as
a proxy of the average size of the jump in the stock price dynamics) and the future stock
return by taking the spread between the ATM call and put option-implied volatilities
(henceforth IV spread) as a measure of the slope of the implied volatility curve. Cremers
& Weinbaum (2010) argue that future stock returns can be predicted by the deviation
from the put-call parity in the equity option market, as stocks with relatively expensive
calls compared to otherwise identical puts earn approximately 50 basis points per week
more in profit than the stocks with relatively expensive puts.
This paper offers several contributions to the existing literature. First, this study
examines whether IV convexity exhibits significant predictive power for future stock
returns even after controlling for the effect of IV slope and other firm-specific
characteristics. Although recent evidence shows that the skewness component of the
risk-neutral distribution of underlying stock returns, our research is, to the best of our
knowledge, the first study that makes a sharp distinction between the 3rd and 4th
moments of equity returns implied by option prices. Specifically, we decompose the
information extracted from the shape of option-implied volatiltiy curve into separate IV
3
slope and IV convexity measures, and empirically verify that IV convexity significantly
predicts the cross section of future stock returns even after controlling for the IV slope
effect. Furthermore, IV spread measure proposed by Yan (2011) simply captures the
effect of the average jump size but not the effect of jump-size volatility in the SVJ
model framework. We extend those findings by examining how IV convexity explains
the cross-section of future stock returns to address the jump-size volatility effect.
Furthermore, this paper overcomes the potential caveat of ex-post information from past
realized returns in the previous studies on the effect of skewness (e.g., Kraus &
Litzenberger (1976); Lim (1989); Harvey & Siddique (2000)) by estimating an ex-ante
measures of skewness (IV slope) and excess kurtosis (IV convexity) from the
forward-looking option price data.6
This paper sheds new light on the relationship between the higher-moment
information from individual equity option prices and the cross-section of future stock
returns. Chang et al. (2013) investigate how market-implied skewness and kurtosis affect
the cross-section of stock returns by looking at the risk-neutral skewness and kurtosis
implied by index option prices based on the model framework proposed by Bakshi,
Kapadia & Madan (2003). However, their approach ignores the idiosyncratic
components of option-implied higher moments in stock returns.7 Our paper extends the
findings of Chang et al. (2013) by employing firm-level equity option price data, and
further decomposing IV convexity into systematic and idiosyncratic components to fully
identify the lead-lag relationship between IV convexity and the cross-section of future
stock returns.
2. Motivation
In this section, we demonstrate the asset pricing implications of our proposed IV
convexity measure. As known, an option-implied risk-neutral distribution of the
underlying stock return exhibits heavier tails than the normal distribution with the same
mean and standard deviation, in the presence of higher moments such as skewness and
excess kurtosis.8 Accordingly, information about these higher moments embedded in the
various shapes of implied volatility curves can be examined from various perspectives.
6The ex-post higher moments estimated from the realized stock returns can be biased unless the returndistribution is stationary and time-invariant; refer to Bali et al. (2015) and Chang et al. (2013).
7Yan (2011) finds that both the systematic and idiosyncratic components of IV spread are priced andthat the latter dominates the former in capturing the cross-sectional variation of future stock returns.
8Hereafter, we use kurtosis and excess kurtosis interchangeably, despite their conceptual differences.
4
Figure 1: Shape of the Implied Volatility Curves
This figure illustrates the effect of different values of skewness and excess kurtosis of underlying
asset returns on the shape of the implied volatility curve. The base parameter set is taken as (r,
σ, skewness, excess kurtosis)=(0.05, 0.3,−0.5, 1.0) where r is the risk-free rate and σ is standard
deviation, and (S0, T ) = (100, 0.5) as an illustrative example.
0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.20.2
0.22
0.24
0.26
0.28
0.3
0.32
0.34
0.36
Moneyness (K/S)
Impl
ied
Vol
atili
ty
Skewness = −1.0Skewness = −0.75Skewness = −0.5Skewness = −0.25Skewness = 0.0
0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.20.23
0.24
0.25
0.26
0.27
0.28
0.29
0.3
0.31
0.32
0.33
Moneyness (K/S)
Impl
ied
Vol
atili
ty
Kurtosis = 0.0Kurtosis = 0.5Kurtosis = 1.0Kurtosis = 1.5Kurtosis = 2.0
2.1. Risk-neutral Higher-order Moments
To explore the effects of risk-neutral skewness and excess kurtosis on option pricing, we
consider a geometric Levy process to model the risk-neutral dynamics of the underlying
stock price given by
St = S0eXt , (1)
where X is the return process whose increments are stationary and independent. In this
context, a natural characterization of a probability distribution is specifying its cumulants,
as we can readily expand the probability distribution function via the Gram-Charlier
expansion, a method to express a density probability distribution in terms of another
(typically Gaussian) probability distribution function using cumulant expansions.9 This
feature aids in understanding how the skewness and kurtosis of the underlying asset return
affect the shape of the option-implied volatility curves.10
Figure 1 illustrates the effect of different values of skewness and excess kurtosis on the
shape of an implied volatility curve. We can observe that a negatively skewed distribution,
ceteris paribus, leads to a steeper volatility smirk, whereas an increase in the excess
9The nth cumulant is defined as the nth coefficient of the Taylor expansion of the cumulant generatingfunction, the logarithm of the moment generating function. Intuitively, the first cumulant is the expectedvalue, and the nth cumulant corresponds to the nth central moment for n = 2 or n = 3. For n ≥ 4, thenth cumulant is the nth degree polynomial in the first n central moments.
10See Tanaka, Yamada & Watanabe (2010) for details of the application of Gram-Charlier expansionfor option pricing.
5
Figure 2: Higher Moments of Underlying Asset Return, IV slope and IV convexity
This figure shows the effect of skewness and excess kurtosis of underlying asset returns on IV
slope and IV convexity. The base parameters are consistent with those of Figure 1. We take
0.8, 1.0 and 1.2 as the moneyness (K/S) points for IV slope and IV convexity.
−1 −0.8 −0.6 −0.4 −0.2 00
0.02
0.04
0.06
0.08
0.1
0.12
0.14
Risk−neutral Skewness
IV SlopeIV Convexity
0 0.5 1 1.5 20
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
Risk−neutral Kurtosis
IV SlopeIV Convexity
kurtosis makes the volatility curve more convex. This observation is consistent with the
theoretical verification of Zhang & Xiang (2008) in that the slope and the curvature of
the implied volatility smirk are asymptotically proportional to the risk-neutral skewness
and excess kurtosis of the underlying asset returns, respectively. We further define the
implied volatility convexity (IV convexity) and the implied volatility slope (IV slope) as
IV convexity = IV (OTM) + IV (ITM)− 2× IV (ATM), (2)
IV slope = |IV (OTM)− IV (ITM)| , (3)
where IV (·) denotes the implied volatility as a function of an option’s moneyness or delta.
Figure 2 confirms the distinct option pricing implications between IV slope and IV
convexity. The 3rd risk-neutral moment of underlying return distribution is negatively
related with IV slope but we find little relationship between the risk-neutral skewness
and IV convexity. On the other hand, IV convexity shows a positive relationship with
the 4th risk-neutral moment of underlying stock return, whereas IV slope reflects little
information on the risk-neutral kurtosis of the underlying equity distribution. This
observation provides the unique opportunity to examine the effect of higher moments
embedded in option prices on the predictability of future stock prices.
We further investigate several asset pricing implications on the risk-neutral
higher-order moments in Appendix A by showing that IV convexity indicates the
risk-neutral excess kurtosis driven by the perceived volatility of stochastic volatility (σv)
6
and jump volatility (σJ) of underlying asset returns. The results confirm that IV
convexity is a forward-looking measure of excess tail-risk contribution to the perceived
variance of underlying equity returns under the risk-neutral measure.
2.2. Hypothesis Development
Having observed that the convexity of an option-implied volatility curve is a
forward-looking measure of the perceived likelihood of extreme movements in the
underlying equity price, we develop our empirical research question by examining how
IV convexity is related to the cross-section of future underlying stock returns. Notice
that the aforementioned theoretical analyses are based on the perfect information flow
between options and stock markets, where expected returns are irrelevant for option
pricing and equity investors express their tail-risk aversion simultaneously when equity
option investors generate leptokurtic implied distributions of future stock returns.
In reality, however, numerous empirical studies have identified that options market
would be an ideal venue for informed trading and options market could lead stock market
in the price discovery process, as informed traders have incentives to trade in options
markets to benefit from their informational advantage. Specifically, higher risk-neutral
excess kurtosis in the options market, ceteris paribus, can be an early warning signal of
the realized tail-risk aversion in the equity market, and vice versa. As such, we expect
that a persistent predictive capability of IV convexity for future stock returns with a
negative relationship would appear to link the risk-neutral excess kurtosis of the option-
implied stock returns to tail-risk aversion of stock investors with a significant time lag.
We summarize our research questions in the following testable main hypothesis:
• Main hypothesis: A slow and one-way information transmission from options to
stock markets enables informed options traders to capitalize on their private
information on the forthcoming of ex-post tail-risk perception in the stock market.
As a result, IV convexity would appear to link the excess kurtosis of the
risk-neutral stock returns to the realization of equity investors’ tail-risk aversion
associated with a higher tail-risk premium (or, equivalently, a larger discount rate)
with a significant time lag. This delayed price discovery mechanism will show a
persistent predictive capability of IV convexity for future stock returns with a
negative relationship.
The overall goal of our empirical analysis is to examine if the IV convexity measure
can show a significant cross-sectional predictive power for future equity returns. If we
7
observe negative relationship between IV convexity and future stock return with statistical
significance, it would empirically support the existing literature demonstrating the lead-
lag relationship between the options and stock markets in terms of the realization of
delayed tail-risk premium in the stock market.
3. Empirical Analysis
This section introduces our data set and methodology to estimate option-implied
convexity in a cross-sectional manner. We test whether IV convexity has strong
predictive power for future stock returns with statistical significance. Then, we compare
the impact of IV convexity with that of option-implied volatility slope on the future
underlying stock returns in the presence of various control variables.
3.1. Data and Sample
We obtain the U.S. equity and index option data from OptionMetrics on a daily basis
from January 2000 through December 2013. As the raw data include individual equity
options in the American style, OptionMetrics applies the binomial tree model of Cox,
Ross & Rubinstein (1979) to estimate the options-implied volatility curve to account for
the possibility of an early exercise with discrete dividend payments. Employing a kernel
smoothing technique, OptionMetrics offers an option-implied volatility surface across
different option deltas and time-to-maturities. Specifically, we obtain the fitted implied
volatilities on a grid of fixed time-to-maturities, (30 days, 60 days, 90 days, 180 days,
and 360 days) and option deltas (0.2, 0.25, . . . , 0.8 for call options and -0.8, -0.75, . . . ,
-0.2 for put options), respectively. Following An et al. (2014) and Yan (2011), we then
select the options with 30-day time-to-maturity at the end of each month to examine the
predictability of IV convexity for future stock returns.
[Table 1 about here.]
Table 1 shows the summary statistics of the fitted implied volatility and fixed deltas
of the individual equity options with one month (30 days) time-to-maturity chosen at the
end of each month. We can observe a positive convexity in the option-implied volatility
curve as a function of the option’s delta in that the implied volatilities from in-the-money
(ITM) options (calls for delta of 0.55 ∼ 0.80, puts for delta of −0.80 ∼ −0.55) options
and OTM (calls for delta of 0.20 ∼ 0.45, puts for delta of −0.45 ∼ −0.20) are greater on
8
average than those near the ATM options (calls for delta of 0.50, puts for delta of −0.50).
Panel B of Table1 reports the unique number of firms by industry each year. Each firm is
assigned into one of the Fama-French 12 industries (FF1-12) classification based on its SIC
code. There are 2,900 unique firms in 2000, rising to 4,159 in 2013. We obtain daily and
monthly individual common stock (shrcd in 10 or 11) returns from the Center for Research
in Security Prices (CRSP) for stocks traded on the NYSE (exchcd=1), Amex (exchcd=2),
and NASDAQ (exchcd=3). Stocks with a price less than three dollars per share are
excluded to weed out very small or illiquid stocks and the potential extreme skewness
effect (Loughran & Ritter (1996)). Accounting data is obtained from Compustat. We
obtain both daily and monthly data for each factor from Kenneth R. French’s website.11
3.2. Variables and portfolio formation
We adopt IV convexity as a risk-neutral proxy for the option-implied tail-risk perception
on the underlying stock returns over the life of option. Throughout this paper, we propose
our main measure of IV convexity defined as
IV convexity = IVput(−0.2) + IVput(−0.8)− 2× IVcall(0.5), (4)
where IVput(∆) and IVcall(∆) refer to the fitted put and call option-implied volatilities
with one month (30 days) to expiration, and ∆ is each options’ delta.12 Note that the
violation of put-call parity is commonly observed in practice, as the parity does not
account for potential market frictions and there is an early exercise possibility for deep
in-the-money American put options; e.g., Yan (2011) and Cremers & Weinbaum (2010)
find that the spread between put- and call-implied volatilities can predict the cross-section
of unerlying stock returns.
As shown in (4), IV convexity is defined as the sum of OTM and ITM put implied
volatilities less double the ATM call implied volatility.13 The rationale is that those who
have private information on the (forthcoming) widespread tail-risk aversion would buy a
cheap deep-OTM put option as a protection against the potential decrease in the stock
return for hedging purposes or buy a deep-ITM put option for taking its early exercise
premium as a leverage to grab a quick profit for speculative purposes to capitalize on
11http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html.12Delta is commonly used as the industry practice of measuring moneyness, as an option’s delta is
sensitive to the option’s intrinsic value and time-value at the same time. Our results are robust acrossdifferent measures of moneyness, as we consider short-term maturity options in our analysis.
13See Section 3.4 for alternative measures of IV convexity.
9
private information.14 For the same reasoning, an informed investor who anticipate
heavier tail-risk aversion prevailing in the stock market would sell call options. We
subtract double the implied volatility of the ATM call option, which is generally the
most frequently traded option best reflecting market participants’ sentiment regarding
the firm’s future condition.
We also consider other measures related to the shape of option-implied volatility
curve. Specifically, option-implied volatility level and slope, denoted by IV level and IV
slope respectively, are defined as
IV level = 0.5× [IVput(−0.5) + IVcall(0.5)] , (5)
IV slope = IVput(−0.2)− IVput(−0.8), (6)
respectively. This decomposition enables us to investigate whether the IV slope and
IV convexity actually have distinct impacts on a cross-section of future stock returns,
thereby to check whether IV convexity carries extra predictability of future stock returns
controlling for return predictability of IV slope. Furthermore, we define IV spread as
IV spread = IVput(−0.5)− IVcall(0.5), (7)
which is proposed by Yan (2011), who argues that the IV spread mainly reflects the
contribution of 3rd moments of return distributions driven by the mean jump size in the
price dynamics. Next, motivated by Xing et al. (2010), we define IV smirk as the difference
between the implied volatility of the OTM put options and the ATM call options on the
same stock. If there are more than one record of OTM put or ATM call options for a
certain stock on the same day, we take the OTM put option with the moneyness (= K/S)
closest to 0.95 and the ATM call option with its moneyness (= K/S) closest to 1.15
14Although an ITM put option is costly, its demand can increase under strong short-sale constraints ofunderlying stocks with pessimistic tail-risk perception, as a severe short-selling constraint would make itmore expensive to capitalize on their negative private information in the equity market, where ITM putoptions would be rational alternatives. Recall that the delta of a deep-ITM American put option is closeto -1 owing to the possibility of early exercise.
15Xing et al. (2010) estimate IV smirk from the implied volatilities of options data, for which we applythe following filters to minimize the impact of recording errors. Following their methodology, we eliminatethe prices that violate no-arbitrage condition by using call data only when the call option prices fall insideof the interval (St −Dt,T −Ke−rt,T , S) and use put data only when the put option prices remain insideof the interval (Ke−rt,T − (St −Dt,T ), S), where S is the price of underlying stock, K is the strike priceof options, r is risk-free rate and D is the dollar dividend and T is the expiration time. We remove theoptions from the sample if their average of best bid price and best ask price is higher than $0.125, thebid price is equal to zero, or the ask price is lower than the bid price. We include the option contractwith positive open interest and non-missing volume data. The time-to-maturity of the options must bebetween 10-60 days and the implied volatility of the option should be between 3% and 200%.
10
Taking advantage of both equity options and index options data, we can disentangle
IV convexity into systematic and the idiosyncratic components. For this purpose, we run
the time series regression each month using the S&P500 index options with 30-day time-to
maturity as a benchmark for the market along with individual equity options at the end
of each month given by
IV convexityi,k = αi + βi × IV convexityS&P500,k + εi,k, (8)
where t − 30 ≤ k ≤ t − 1 on a daily basis. We define the fitted values and residual
terms as the systematic component of IV convexity (convexitysys) and the idiosyncratic
component of IV convexity (convexityidio), respectively.
[Table 2 about here.]
At the end of each month, we compute the cross-sectional IV level, IV slope, IV
convexity, IV smirk, and IV spread measures from 30-day time-to-maturity options. Panel
A of Table 2 shows the descriptive statistics for each implied volatility measure computed
at the end of each month using 30-day time-to-maturity options. As for the average values
for each of variable, IV level has 0.4739, IV slope for 0.0423, IV spread for 0.0090, IV
smirk for 0.0687, and IV convexity for 0.0952, respectively. The standard deviation of
IV convexity is 0.2774, and 0.1703 for convexitysys, 0.2235 for convexityidio, respectively.
Panel B of Table 2 reports the descriptive statistics of firm characteristics and asset pricing
factors.16
3.3. Predictive power of IV convexity
We now report our empirical results regarding the predictive power of IV convexity for
the cross-section of future underlying stock returns. When constructing a single sorted IV
convexity portfolio, we sort all stocks at the end of each month based on the IV convexity
and match with the subsequent monthly stock returns. The IV convexity portfolios are
rebalanced every month.
[Table 3 about here.]
Panel A of Table 3 reports the means and standard deviations of the five IV convexity
quintile portfolios and average monthly portfolio returns of both equal-weighted (EW) and
16See Appendix B for our variable definitions of firm characteristics (Size, BTM) and asset pricingfactors (Market β, MOM, REV, ILLIQ and Coskew).
11
value-weighted (VW) portfolios over the entire sample period. To examine the relationship
between IV convexity and future stock returns, we form five portfolios according to the
IV convexity value at the end of each month. Quintile 1 is composed of stocks with the
lowest IV convexity, while Quintile 5 is composed of stocks with the highest IV convexity.
These portfolios are rebalanced every month, and assumed to be held for the subsequent
one-month period. As shown, the average EW portfolio return monotonically decreases
from 0.0208 for the lowest quintile portfolio Q1 to 0.0074 for the highest quintile portfolio
Q5. The average monthly return of the arbitrage portfolio buying the lowest IV convexity
portfolio Q1 and selling highest IV convexity portfolio Q5 is significantly positive (0.0134
with t-statistics of 7.87). The average VW portfolio returns exhibit a similar decreasing
pattern from Q1 (0.0136) to Q5 (0.0023), and the return of zero-investment portfolio
(Q1-Q5) is significantly positive (0.0113 with t-statistics of 5.08).
Panel B and C of Table 3 show the average monthly returns of quintile portfolios
sorted by IV spread and IV smirk, respectively. The EW portfolios sorted by IV spread
show that their EW portfolio returns decrease monotonically from 0.0145 for quintile
portfolio Q1 to 0.0013 for quintile portfolio Q5, where the average return difference
between Q1 and Q5 amounts to 0.0131 with t-statistics of 7.31 and similar patterns are
observed with VW portfolios sorted by IV spread. These results certainly confirm the
empirical finding of Yan (2011) in that low IV spread stocks outperform high IV spread
stocks. In a similar vein, we find from Panel C that the average returns of quintile
portfolios sorted by IV smirk are decreasing in IV smirk, and the returns of
zero-investment portfolios (Q1-Q5) are all positive and statistically significant for both
the EW and VW portfolios, consistent with Xing, Zhang and Zhao (2010) in that there
exists a negative predictive relationship between IV smirk and future stock return.
Left panel of Figure 3 shows the monthly average IV convexity value for each quintile
portfolio, while Right panel plots the monthly average return of the arbitrage portfolio
formed by taking long position in the lowest quintile and short position in the highest
quintile portfolios (Q1-Q5). The time-varying average monthly returns of the long-short
portfolio are mostly positive, confirming the results reported in Table 3.
3.4. Alternative IV convexity measures
We next consider alternative measures of options-implied volatility convexity than ours
given by (4). We first explore various IV convexity measures by varying ITM and OTM
points of put options. Panel A of Table 4 reports descriptive statistics for average portfolio
returns based on the main IV convexity measure given by (4) estimated with various delta
12
Figure 3: Average IV convexity and Quintile Portfolio Returns
This figure shows the time-series behavior of the average IV convexity of the quintile portfolios
(Left panel) and the monthly average returns of the long-short portfolios Q1-Q5 (Right panel)
from Jan 2000 to Dec 2013 on a monthly basis.
Jan-00 01-Sep 03-May 05-Jan 06-Sep 08-May 10-Jan 11-Sep 13-May-0.5
0
0.5
1
1.5
Q1Q2Q3Q4Q5
Feb-00 01-Oct 03-Jun 05-Feb 06-Oct 08-Jun 10-Feb 11-Oct 13-Jun-0.05
0
0.05
0.1
0.15
Q1-Q5
points. The decreasing patterns in portfolio returns are still observed across different delta
points and arbitrage portfolio (Q1-Q5) returns are significantly positive for both the
equal-weighted and value-weighted portfolio returns. This result confirms the negative
relationship between IV convexity and future stock returns whatever delta points of put
options we use to compute convexity.
[Table 4 about here.]
Next, we construct various alternative measures of IV convexity, denoted by
IV convexity(·), between calls and puts given by
IV convexity(II) = IVput(−0.2) + IVput(−0.8)− 2× IVput(−0.5), (9)
IV convexity(III) = IVcall(0.2) + IVcall(0.8)− 2× IVcall(0.5), (10)
IV convexity(IV ) =IV convexity(II) + IV convexity(III)
2, (11)
IV convexity(V ) = IVput(−0.2) + IVcall(0.2)− 2× IVput(−0.5), (12)
IV convexity(V I) = IVput(−0.2) + IVcall(0.2)− 2× IVcall(0.5). (13)
As a benchmark, we first construct a put-based convexity measure, IV convexity(II), and
a call-based measure, IV convexity(III). Then, we define IV convexity(IV ) as the average
of IV convexity(II) and IV convexity(III) to incorporate comprehensive implied volatility
information from call and put options. Further, IV convexity(V ) and IV convexity(V I) are
13
constructed by the sum of OTM option-implied volatilities less the ATM option-implied
volatility. The rationale behind them is the fact that OTM put (call) options are generally
more liquid than the ITM call (put) options at the same strike prices.
We then construct a set of single-sorted quintile portfolios (from Q1 to Q5) by
sorting all stocks at the end of each month based on each of the alternative IV convexity
measures and match with the subsequent monthly stock returns. Quintile 1 is composed
of stocks with the lowest IV convexity(·) while Quintile 5 is composed of stocks with the
highest IV convexity(·). That is, at the end of each month, all firms are assigned to one
of five portfolio groups based on alternative options implied convexity. These portfolios
are equally weighted, rebalanced every month, and assumed to be held for the
subsequent one-month period. This process is repeated in every month. Panel B of
Table 4 reports the descriptive statistics of the average portfolio returns sorted by
alternative measures of IV convexity given by (4)-(13). Our empirical results show that
the alternative IV convexity measures generally produce a monotone decreasing pattern
of the average quintile portfolio returns, and the realized returns of the arbitrage
portfolio (Q1-Q5) across different definitions of IV convexity remain positive with
statistical significance.17 This result confirms our main hypotheses in that the negative
relationship between IV convexity and future stock returns are robust and consistent
across different definitions of IV convexity.
3.5. Controlling Firm-specific Factors
We investigate whether the positive arbitrage portfolio returns (Q1-Q5) are still significant
even after controlling for firm-specific equity risk factors. Accordingly, we test whether a
set of representative firm characteristics crowd out the negative relationship between IV
convexity and stock returns. To examine whether the relationship between IV convexity
and stock returns disappear after controlling for the firm characteristics, we double-sort
all stocks following Fama & French (1992). Specifically, all stocks are sorted into five
quintiles by ranking on standard equity risk factors and then sorting within each quintile
into five quintiles according to IV convexity.
[Table 5 about here.]
We first adopt the suggestion of Fama & French (1993) in that firm size,
book-to-market ratio, and Market β are regarded as the three main firm characteristic
17The mean arbitrage portfolio returns are significantly positive at 5% confidence level across differentmeasures of IV convexity except for the case of IV convexity(II) at 10% level of confidence.
14
factors of stock returns. Table 5 reports the average monthly returns of the 25 (5 × 5)
portfolios sorted first by firm characteristic risks (firm size, book-to-market ratio, and
Market β) and then by IV convexity and average monthly returns of the long-short
arbitrage portfolios (Q1-Q5). We can observe that the average monthly portfolio returns
generally decline as the average firm-size increases. As shown in the results from
double-sorting using firm-size and IV convexity, we find that the returns of the IV
convexity quintile portfolios are still decreasing in IV convexity in most size quintiles,
and the return of all zero-investment portfolios (Q1-Q5) in size quintiles are all positive
and statistically significant. In particular, the positive difference in the smallest quintile
is the largest (0.0187) compared to the other size quintile portfolios.
The two-way cuts on book-to-market and IV convexity show that the higher book-
to-market portfolio gets more returns compared to the lower book-to-market portfolios
in each IV convexity quintile. The decreasing patterns in IV convexity portfolio returns
remain even after controlling the systematic compensation drawn from the book-to-market
factor. Note that the overall zero-cost portfolios formed by long Q1 and short Q5 are also
positive and statistically significant: 0.0097 (t-statistic of 4.82) for B1 (BTM quintile 1),
0.0121 (t-statistic of 5.76) for B2, 0.0108 (t-statistic of 5.42) for B3, 0.0134 (t-statistic of
5.67) for B4, and 0.0177 (t-statistic of 5.19) for B5.
When sorting the 25 portfolios first by Market β and then by IV convexity, the
negative relationship between IV convexity and stock return persists, implying that this
decreasing pattern cannot be explained by Market β. Note that the average monthly
portfolio returns generally increase as the average Market β rises.
We also consider the other four pricing factors of (i) the momentum effect documented
by Jegadeesh & Titman (1993), (ii) the short-term reversal suggested by Jegadeesh (1990)
and Lehmann (1990), (iii) the illiquidity proposed by Amihud (2002) and (iv) co-skewness
suggested by Harvey & Siddique (2000) to examine whether the decreasing pattern of
portfolio returns in IV convexity disappears when controlling these systematic risk factors.
Stocks are first sorted into five groups based on their momentum (or reversal, illiquidity,
co-skewness) measures and then sorted by IV convexity forming 25 (5 × 5) portfolios.
[Table 6 about here.]
Table 6 presents the returns of 25 portfolios sorted by each of four equity pricing
factors (momentum, reversal, illiquidity, co-skewness) and IV convexity. When we look
at the momentum patterns in the momentum-IV convexity portfolios, winner portfolios
consistently achieve more abnormal returns than loser portfolios except for the lowest
15
momentum-lowest IV convexity quintiles, which could be caused by using a different
sample datasets compared to that in Jegadeesh & Titman (1993). While Jegadeesh &
Titman (1993) use only stocks traded on the NYSE (exchcd=1) and Amex (exchcd=2),
we add stocks traded on the NASDAQ (exchcd=3). For the holding period strategies,
Jagadeesh and Titman (1993) adopt 3-, 6-, 9-, and 12-month holding periods, while this
study assumes that portfolios are held for one month.
Even after controlling momentum as a systematic risk, we observe that the portfolio
return differential between the lowest and highest IV convexity in each momentum
quintile remains significantly positive, indicating that IV convexity contains
economically meaningful information that cannot be explained by the momentum factor.
For the reversal-IV convexity double sorted portfolio case, there is a clear reversal
patterns in most cases when using a reversal strategy (i.e., the past winner earns higher
returns in the next month compared to past loser), though there are some distortions
in the lowest reversal-highest IV convexity quintiles. The Q1-Q5 strategy of buying and
selling stocks based on IV convexity in each reversal portfolio and holding them for one
month still earns significantly positive returns. This implies that the same results still
hold even after controlling for reversal effects.
We further incorporate the Amihud (2002) measure of illiquidity to address the role
of the liquidity premium in asset pricing. Amihud (2002) finds that the expected market
illiquidity has positive and highly significant effect on the expected stock returns, as
investors in the equity market require additional compensation for taking liquidity risk.
We examine whether the market illiquidity measure (ILLIQ) explains the higher return
on the lowest IV convexity stock portfolio (Q1) relative to the highest IV convexity
stock portfolio (Q5). The double-sorted quintile portfolios by ILLIQ andIV convexity
exhibit analogous patterns in that their average returns tend to decrease in IV
convexity. The zero-investment portfolios (Q1-Q5) based on the ILLIQ quintiles
demonstrate their significantly positive average returns across different ILLIQ quintiles.
This finding implies that a significantly negative IV convexity premium remains even
after we control for the illiquidity premium effect.
Next, we consider conditional skewness, as Harvey & Siddique (2000) find that
conditional skewness (Coskew) can explain the cross-sectional variation of expected
returns even after controlling factors based on size and book-to-market value. To
examine whether this Coskew factor captures the higher returns of the lowest IV
convexity stocks relative to the highest IV convexity stocks, we constructed 25 portfolios
sorted first by Coskew and then by IV convexity. For the Coskew-IV convexity double
16
sorted portfolios, there is a clear decreasing pattern in IV convexity within each Coskew
sorted quintile portfolio. The returns of zero-cost IV convexity portfolios (Q1-Q5)
within each Coskew quintile portfolio are all positive and statistically significant: 0.0130
(t-statistic of 4.60) for C1 (Coskew quintile 1), 0.0090 (t-statistic of 3.83) for C2, 0.0040
(t-statistic of 2.39) for C3, 0.0076 (t-statistic of 4.18) for C4, and 0.0127 (t-statistic of
5.07) for C5, respectively. A negative IV convexity premium remains significantly even
after we control for the Coskew premium effect. After all, we conclude that the negative
relationship between IV convexity and stock return consistently persists even after
controlling for various kinds of firm-specific risks identified in prior researches.
3.6. Correcting for implied-volatility skew
It is empirically important to figure out whether the implication of IV convexity is really
significant even after we correct for the skew of the implied-volatility curves. To answer
this question, we examine if the negative relationship between IV convexity and stock
returns persists after controlling for the effect of the option-implied volatility slope on
stock returns suggested by other researchers. For this purpose, we consider the following
three measures for the slope of the option-implied volatility curve: (i) IV slope, (ii) IV
spread, and (iii) IV smirk, where the last two measures are defined and estimated as in
Yan (2011) and Xing et al. (2010), respectively.
[Table 7 about here.]
Table 7 presents the average monthly returns of 25 portfolios sorted first by IV
slope, IV spread, andIV smirk and then sorted by IV convexity within each of IV slope,
IV spread, and IV smirk sorted quintile portfolios, respectively. The first five columns
show the results of our IV slope-IV convexity double sorted portfolio returns. We can
observe a decreasing pattern with respect to IV convexity in each IV slope quintile, and
the Q1-Q5 strategy based on IV convexity in each IV slope portfolio and holding them
for one month returns significantly positive profits across the different specifications. As
for IV spread, the IV convexity strategy that buys the lowest quintile portfolio and sells
the highest quintile portfolio within each IV spread portfolio yields significantly positive
returns in all cases, suggesting that IV spread does not capture the IV convexity effect.
It is also noteworthy that IV convexity arbitrage portfolios (Q1-Q5) from S1 to S5 IV
smirk portfolios produce significantly positive returns. This observation reveals that IV
smirk cannot fully capture the negative relationship between IV convexity and future
stock returns across different IV smirk quintile portfolios.
17
It might be the case that the predictive power of IV convexity is partially driven by
IV smirk and vice versa. To disentagle the problem, we decompose IV convexity into the
IV smirk part and the orthogonalized IV convexity part by regressing daily IV convexity
against IV smirk. Specifically, the orthogonalized implied volatility convexity, denoted by
orthogonalized IV convexity, is defined as the residual terms of the regression specification
given by
IV convexityi,k = αi + βi × IV smirki,k + εconvexityi,k , (14)
where t− 30 ≤ k ≤ t− 1 on a daily basis. In a similar vein, we define the orthogonalized
IV smirk as the residual terms of the regression specification given by
IV smirki,k = αi + βi × IV convexityi,k + εsmirki,k . (15)
We then sort all stocks at the end of each month based on the orthogonalized IV
convexity (εconvexity) and orthogonalized IV smirk (εsmirk), and match with the subsequent
monthly stock returns by rebalancing the 5 quintile (from the lowest Q1 to the highest Q5)
portfolios every month. As a result, we observe that the average portfolio return decreases
from Q1 (0.0090) to Q5 (0.0064), and the monthly return differential between Q1 and Q5
is 0.0025 with t-statistics of 2.74. This result indicates that the return predictability of IV
convexity cannot be fully explained by that of IV smirk. On the other hand, it is shown
that the predictive power of orthogonalized IV smirk is statistically insignificant, as the
difference between average return of Q1 (0.0085) and that of Q5 (0.0084) is less than
0.0001 with t-statistics of 0.030. That is, the return predictability of IV smirk disappears
if we correct for the effect of IV convexity.
All in all, these findings support the proposition that the negative relationship of
IV convexity to future stock returns still holds after considering the impact of IV slope,
IV spread, and IV smirk on stock returns. Our observation implies that the impact of
IV convexity on the distribution of stock returns does not fully come from IV slope, IV
spread, and IV smirk and that IV convexity is an important factor in determining the
fat-tailed distribution characteristics of future stock returns.
3.7. Systematic and Idiosyncratic Components of IV convexity
Total risk of stock returns can be decomposed into systematic and idiosyncratic
components. In theory, the systematic risk factor (Market β) should be priced in
equilibrium, while the idiosyncratic risk factor cannot capture the cross-sectional
18
variation of stock returns. However, the real-world investors cannot perfectly diversify
away the idiosyncratic risks, thereby it is often observed that idiosyncratic risk can also
play an important role in the cross-sectional variation of stock returns. In this context,
we decompose IV convexity into systematic and idiosyncratic components to further
investigate the source of negative relationship between IV convexity and the
cross-section of future stock returns. As our data set includes both equity options and
index options, we can effectively decompose IV convexity into the systematic and
idiosyncratic components by performing a simple regression analysis given by (8).
[Table 8 about here.]
Panel A of Table 8 provides the descriptive statistics for the average portfolio
returns sorted by systematic components and idiosyncratic components of IV convexity.
It shows that the average portfolio return monotonically decreases from Q1 to Q5 and
that the return differential between Q1 and Q5 is significantly positive. It is worth
noting that the negative pattern is robust even if we decompose IV convexity into the
systematic and idiosyncratic components. That is, convexitysys and convexityidio reveal
decreasing patterns in the portfolio returns as the IV convexity portfolio increases. The
difference between the lowest and the highest quintile portfolios sorted by convexitysys
and convexityidio are significantly positive with t-statistics of 6.56 and 5.72, respectively.
This implies that both components have predictive power for future portfolio returns
and are significantly priced.
Panel B of Table 8 reports the average monthly portfolio returns of the 25 quintile
portfolios formed by sorting stocks based on (convexitysys, convexityidio) first, and then
sub-sorted by IV convexity in each of convexitysys or convexityidio quintile. This will
allow us to figure out how the systematic or idiosyncratic components contribute to IV
convexity. In other words, if the decreasing patterns of returns in IV convexity portfolio
become less clearly observed under the control of (convexitysys, convexityidio), this can
be interpreted as a component of convexitysys or convexityidio and can mostly explain the
cross-sectional variation of return on IV convexity compared to the other component of
convexitysys or convexityidio .
As for the results from the sample sorted first by convexitysys and then by IV
convexity, the decreasing patterns generally persist for the convexitysys quintiles, though
there are some distortions in the 1st and 3rd convexitysys quintiles for the highest IV
convexity quintiles. Note that the arbitrage portfolio’s returns (Q1-Q5) in each
convexitysys quintile portfolio still remain large and statistically significant. As for the
19
convexityidio-IV convexity double sorted portfolios shown in the right-hand side of Table
8, the negative relationship between IV convexity and average portfolio return exists,
though the order of portfolio returns are not perfectly preserved in the 3rd and 5th
convexityidio quintiles. In addition, the long-short IV convexity portfolio returns in the
convexityidio quintile portfolio (Q1-Q5) are significantly positive with a t-statistic larger
than two. Thus, these results provide evidence that neither component can fully capture
and explain all cross-section variations of returns on IV convexity, but both components
(convexitysys and convexityidio) have decreasing patterns of portfolio returns, and are
needed to capture the cross-sectional variations of returns.
3.8. Short-selling Constraints and Information Asymmetry
We next look at the negative relationship between IV convexity and future stock returns
with respect to firm-level information asymmetry and short-selling constraints. We first
hypothesize that the decreasing pattern of the IV convexity portfolio returns becomes
more pronounced for the firms with more restrictive short-selling constraints. Secondly,
we infer that informed traders have more opportunities to get higher profit from the stocks
with more severe disparity of information possessed by informed traders. Subsequently,
the decreasing pattern of the IV convexity portfolio returns will become more pronounced
for the stocks with stronger information asymmetry.
In this context, we employ the measures of analyst coverage and analyst forecast
dispersion as proxies for information asymmetry along with the share of institutional
ownership as a proxy for the short-selling constraints. According to Diether, Malloy &
Scherbina (2002), analyst forecast dispersion is measured by the scaled standard
deviation of I/B/E/S analysts’ fiscal quarterly earnings per share forecasts. Stronger
information asymmetry is implied by less numbered analyst coverage and greater
analyst forecast dispersion. As the measure of short-sale constraint, following Campbell,
Hilscher & Szilagyi (2008) and Nagel (2005), we calculate the share of institutional
ownership by summing the stock holdings of all reporting institutions for each stock on
a quarterly basis. Nagel (2005) refers that short-sale constraints are most likely to bind
among the stocks with high individual (i.e., low institutional) ownership, so the stocks
with less institutional ownership suffer from more binding short-sale constraints.
[Table 9 about here.]
Table 9 reports the average monthly returns of double-sorted portfolios, first sorted
by the previous quarterly percentage of shares outstanding held by institutions obtained
20
from the Thomson Financial Institutional Holdings (13F) database, previous quarter’s
analyst coverage obtained from I/B/E/S, previous quarter’s analyst forecast dispersion
obtained from I/B/E/S, respectively and then sub-sorted according to IV convexity within
each quintile portfolio on a monthly basis. We observe that the decreasing pattern in the
average monthly portfolio returns appears to be more pronounced for the stocks with
institutional investors owning a small fraction of the firm’s share. Furthermore, we find
that the IV convexity zero-cost portfolio within lowest institutional ownership portfolio
earns greater positive return than that within the highest institutional ownership portfolio.
The IV convexity effect appears to be stronger when less sophisticated investors (i.e.,
more individual investors) own a large fraction of a firm’s shares. This result confirms
the hypothesis that the decreasing patterns of the IV convexity portfolio returns become
more pronounced for the stocks with stronger short-selling constraints.
The next five columns show how the degree of information asymmetry affects the
predictability of IV convexity. We find that the decreasing pattern of the IV convexity
quintile portfolio returns becomes stronger for in lower analyst coverage quintiles, albeit
some distortion in AC4 and AC5 quintiles. The last pair of columns show the results based
on the analyst forecast dispersion as a proxy for information asymmetry measure. We find
that the IV convexity effect becomes stronger for high analyst forecast dispersion firms,
and the returns of zero-investment IV convexity portfolios (Q1-Q5) within the lowest AD
quintile portfolio return is 0.0081, while that of the highest AD quintile portfolio return
amounts to 0.0162 on a monthly basis. These results imply that the stocks with strong
information asymmetry and with severe short-sale constraints induce the informed traders
mainly trade in the options market rather than in the stock market. This finding confirms
our Hypothesis in the sense that there is a slow one-way information transmission from
the options market to the stock market owing to the cross-market information asymmetry.
4. Robustness Checks
This section addresses additional aspects of IV convexity for robustness. We first explore
time-series analysis along with different forecasting horizons, and then conduct a Fama-
MacBeth regression analysis with various control variables. Subsequently, we perform a
sub-period analysis as a robustness check.
21
4.1. Time-series Analysis
To test whether the existing risk factor models can absorb the observed negative
relationship between IV convexity and future stock returns, we conduct a time-series
test based on CAPM and the Fama & French (1993) three factor model (FF3). The
Fama & French (1993) model has three factors: (i) Rm − Rf (the excess return on the
market), (ii) SMB (the difference in returns between small stocks and big stocks) and
(iii) HML (the difference in returns between high book-to-market stocks and low
book-to-market stocks). Along with the FF3 model, we also use an extended four-factor
model (FF4) of Carhart (1997) by including a momentum factor (UMD) suggested by
Jegadeesh & Titman (1993).
[Table 10 about here.]
Table 10 reports the coefficient estimates of CAPM, FF3, and FF4 time-series
regressions for monthly excess returns on five portfolios sorted by IV convexity and its
systematic and idiosyncratic components. The left-most six columns are the results
using a portfolio sorted by IV convexity. When running regressions using CAPM, FF3,
and FF4, we generally observe strictly positive intercepts, which are statistically
significant and the quintile portfolio returns show negative patterns when they are
formed by IV convexity. In addition, the differences in the intercept between the lowest
and highest IV convexity are 0.0132 (t-statistic of 7.72) for CAPM, 0.0134 (t-statistic of
7.75) for FF3, and 0.0136 (t-statistic of 7.61) for FF4. Adopting Gibbons, Ross &
Shanken (1989), we test the null hypothesis that all estimated intercepts are
simultaneously zero. We find that the null hypothesis is rejected with a p-value less than
0.001 in the CAPM, FF3, and FF4 model specifications. These results imply that the
widely-accepted existing factors (Rm −Rf , SMB, HML, UMD) cannot fully capture and
explain the negative portfolio return patterns sorted by IV convexity. Thus, we confirm
that IV convexity can capture the cross-sectional variations in equity returns not
explained by existing models such as CAPM, FF3, and FF4.
When we conduct time-series test using the portfolios sorted by decomposed
components of IV convexity into convexitysys and convexityidio, most of the estimated
intercepts are significantly positive in general. The joint test to examine whether the
model explains the average portfolio returns sorted by each component of IV convexity
show that the null hypothesis of zero intercepts from Q1 to Q5 is strongly rejected with
a p-value less than 0.001 in the CAPM, FF3, and FF4 for both systematic and
idiosyncratic components of IV convexlty. Therefore, regardless of whether IV convexity
22
Figure 4: Monthly returns formed by IV convexity with different forecasting horizons
This figure shows the monthly average returns and risk-adjusted returns (implied by Fama-
French 3 and 4 factor models) of the long-short IV convexity portfolios for various forecasting
horizons. “Q1-Q5” plots the average monthly returns of the long short IV convexity portfolio
during Jan 2000 to Dec 2013, whereas αFF3 and αFF4 plot the risk-adjusted (implied by Fama-
French 3 and 4 factor models) monthly returns of the long-short IV convexity portfolios for the
same period. The 95% and 99% confidence intervals are reported in each panel.
2 4 6 8 10 12−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6Q1−Q5
Mon
thly
Por
tfolio
Ret
urn
(in p
erce
nt)
2 4 6 8 10 12−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
αFF3
Forecasting Horizon (in months)
2 4 6 8 10 12−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
αFF4
99% Confidence Interval95% Confidence Interval
is invoked by the market (systematic) or by idiosyncratic risk, none of the components
can be explained by existing standard equity risk factors. These results provide strong
evidence supporting our Hypothesis.
We next examine how long IV convexity effect persists and whether this IV convexity
effect caused by slow information transmission between options market and stock market
disappears as the forecasting horizon increases by investigating the long-short arbitrage
strategy based on IV convexity portfolios. Figure 4 plots the average monthly returns and
the risk-adjusted returns implied by Fama-French 3 and 4 factor models of the long-short
IV convexity portfolios for various forecasting horizons along with their 95% and 99%
confidence intervals. We observe that the average monthly returns of Q1-Q5 portfolio
and the risk-adjusted returns implied by Fama-French 3 and 4 factor models (αFF3 and
αFF4) dramatically decrease (from 1.134 to 0.0031 for Q1-Q5, 0.018 to 0.0025 for αFF3, and
0.012 to 0.0026 for αFF4) during the first two months after the portfolio formation time.
Thereafter, the trading strategy based on IV convexity does not generate economically
meaningful profits. The wide confidence intervals indicate that there is quite little chance
to get positive profits based on the IV convexity information in a long run. Our finding
implies that the arbitrage profits based on the IV convexity information can be realized
in the first few months only, because the arbitrage opportunity from the inter-market
information asymmetry disappears as the forecasting horizon increases.
23
4.2. Fama-MacBeth Regression
The time-series test results indicate that the existing factor models may not be able to
perfectly capture the return predictability of IV convexity. To test if IV convexity can
predict stock returns, we conduct Fama & MacBeth (1973) cross-sectional regressions at
the firm level to investigate whether IV convexity has sufficient explanatory power beyond
others suggested in previous literature. Specifically, we consider Market β estimated by
Fama & French (1992), size (ln mv), book-to-market (btm), momentum (MOM), reversal
(REV), illiquidity (ILLIQ), option-implied volatility skew (IV slope), idiosyncratic risk
(idio risk), implied volatility level (IV level), systematic volatility (v2sys), and idiosyncratic
implied variance (v2idio) as common measures of risks that explain stock returns.18 We run
the monthly cross-sectional regression of individual stock returns of the subsequent month
on IV convexity and other control variables as above.
[Table 11 about here.]
Table 11 reports the averages of the monthly Fama & MacBeth (1973) cross-sectional
regression coefficient estimates for individual stock returns with Market β and other widely
accepted risk factors as a control variable along with the Newey & West (1987) adjusted t-
statistics for the time-series average of coefficients with a lag of 3. The column of Model 1
shows the results with Market β and other stock fundamentals including firm-size (ln mv)
and book-to-market ratio (btm) as control variables. We observe that the coefficient on
IV convexity is significantly negative, and this result confirms our previous finding in the
portfolio formation approach. When we include both IV convexity and IV slope, as shown
in Model 2, the coefficient on IV convexity is still significantly negative, indicating that
IV convexity has a strong explanatory power for stock returns that IV slope cannot fully
capture. The significantly negative coefficients on log(MV) confirm the existence of size
effects shown in earlier studies, whereas the coefficients on btm are significantly positive,
supporting the existence of a value premium.
Models 3 and 4 represent the result using Market β, ln mv, btm, MOM, REV,
ILLIQ, and idiosyncratic risk. These variables are widely accepted stock characteristics
that can capture the cross-sectional variation in stock returns. However, the result is
surprising in that the coefficients on Market β are insignificant, while ln mv and btm
18We do not include the co-skewness factor in the specification of Fama & MacBeth (1973) regression.Harvey & Siddique (2000) argue that co-skewness is related to the momentum effect, as the low momentumportfolio returns tend to have higher skewness than high momentum portfolio returns. Thus, we excludeco-skewness from the Fama-MacBeth regression specification to avoid the multi-collinearity problem withthe momentum factor.
24
have significantly negative and positive coefficients, respectively. The estimated
coefficients on MOM and ILLIQ lose their statistical significance, whereas REV has
significantly negative coefficients. Moreover, the estimated coefficients on idiosyncratic
risk (idio risk) suggested by Ang, Hodrick, Xing & Zhang (2006) are significantly
negative. In an ideal asset pricing model that fully captures the cross-sectional variation
in stock return, idiosyncratic risk should not be significantly priced. Fu (2009) finds a
significantly positive relationship between idiosyncratic risk and stock returns, and Bali
& Cakici (2008) show no significant negative relationship, but insignificant positive
relationships when they form equal-weighted portfolios. However, the statistical
significance of the estimated coefficients on the idiosyncratic risk in Model 3 to Model 4
imply that idiosyncratic risk is negatively priced in the presence of other control
variables. Note that IV convexity has a significantly negative coefficient in the presence
of other control variables including idiosyncratic risk in Model 3. We conjecture that IV
convexity explains the cross-sectional variation in returns that cannot be fully explained
by Market β, ln mv, btm, MOM, REV, ILLIQ or idiosyncratic risk. The statistical
significance of IV convexity remains, even after including IV slope in Model 4.
When alternative ex-ante volatility measures such as IV level, systematic volatility
(v2sys), and idiosyncratic implied variance (v2
idio), are included, as in Models 5-8, the sign
and significance for the IV convexity coefficients remain unchanged. This observation
confirms the negative relationship between IV convexity and future stock returns. All
in all, it can be inferred that there is no evidence that existing risk factors and firm
characteristics suggested by prior research can explain the negative return patterns in
IV convexity, and IV convexity captures the cross-sectional variations in the future stock
returns not explained by existing models.
4.3. Sub-period Analysis
We further examine whether the predictive power of IV convexity depends on the state of
the economic condition. For this purpose, we classify states of the economy based on the
Chicago FED National Activity Index (CFNAI) and the National Bureau of Economic
Research (NBER) recession dummy variable.
[Table 12 about here.]
Table 12 shows the average monthly returns of the Q1-Q5 portfolios sorted by IV
convexity based on the CFIAN and the NBER recession dummy, respectively. In the left
25
panel, we divide the entire sample period into the expansion and contraction periods by
taking the median value of the CFNAI as the threshold level. As shown, the decreasing
patterns of the portfolio returns sorted by IV convexity are consistently observed in
both periods. Furthermore, we find that the Q1-Q5 zero-cost portfolio formed on IV
convexity earns significantly positive return in each sub-period. Interestingly, the values
of average portfolio returns of each quintile in the contraction period are higher than
those in the expansion period. The return of zero-investment IV convexity portfolios
(Q1-Q5) is 0.0074 with the t-statistics of 4.92 for the expansion period, whereas the
return becomes 0.0194 with the t-statistics of 6.65 for the contraction period. Overall,
the trading strategy based on IV convexity seemingly earns more pay-off in the
contraction period than in expansion period, as investors tend to overreact to bad news
and the inter-market information asymmetry exacerbates in the recession state.
In the right panel, we take the NBER business cycle dummy variable to classify
the entire period into the expansion and contraction periods. Specifically, the NBER
recession dummy variable takes the value of one if the U.S. economy is in recession as
determined by the NBER, and vice versa. It is notable that similar results are observed,
as both sub-periods show the decreasing patterns in IV convexity portfolios along with
strictly positive returns in zero-cost portfolios with statistical significance. The decreasing
patterns in the IV convexity portfolios are more pronounced in the contraction period than
those in expansion periods, and the magnitude of the zero-investment (Q1-Q5) portfolio
return in the contraction period is larger than that in the expansion period.
5. Conclusion
This study finds empirical evidence that IV convexity, our proposed measure for the
convexity of an option-implied volatility curve, has a negative predictive relationship
with the cross-section of future stock returns, even after controlling for the slope of an
option-implied volatility curve discussed in recent literature. We demonstrate that the
IV convexity measure, as a proxy of both the volatility of stochastic volatility and the
volatility of stock jump size, reflects informed options traders’ anticipation of the excess
tail-risk contribution to the perceived variance of the underlying equity returns.
Consistent with earlier studies, our empirical findings indicate that options traders
have an information advantage over stock traders in that informed traders anticipating
heavier tail risk proactively choose the options market to capitalize on their private
information. The average portfolio returns sorted by IV convexity monotonically
26
decrease from the lowest (Q1) to highest (Q5) IV convexity quintile portfolios on a
monthly basis, implying that the average monthly return on the arbitrage portfolio
buying Q1 and selling Q5 is significantly positive. This pattern persists after
decomposing IV convexity into systematic and idiosyncratic components. In addition,
the negative relationship between IV convexity and future stock returns is robust after
controlling for various firm-specific risk factors suggested in earlier studies and the
impact of the slope of option-implied volatility curve and other well-documented implied
volatility skew measures. This consistency implies that IV convexity is an important
measure to capture the fat-tailed characteristics of stock return distributions in a
forward-looking manner, leading to the leptokurtic implied distributions of underlying
stock returns before equity investors show their tail-risk aversion.
References
Amihud, Y. (2002), ‘Illiquidity and stock returns: Cross-section and time-series effects’,
Journal of Financial Markets 5, 31–56.
An, B.-J., A. Ang, T. G. Bali & N. Cakici (2014), ‘The joint cross section of stocks and
options’, Journal of Finance 69, 2279–2337.
Ang, A., R. Hodrick, Y. Xing & X. Zhang (2006), ‘The cross-section of volatility and
expected returns’, Journal of Finance 61, 259–299.
Arrow, K. J. (1971), Essays in the Theory of Risk-Bearing, Markham Economics Series,
Amsterdam: North-Holland.
Bakshi, G., C. Cao & Z. Chen (1997), ‘Empirical performance of alternative option pricing
models’, Journal of Finance 52, 2003–2049.
Bakshi, G., N. Kapadia & D. Madan (2003), ‘Stock return characteristics, skew laws,
and differential pricing of individual equity options’, Review of Financial Studies
16, 101–143.
Bali, T. G. & A. Hovakimian (2009), ‘Volatility spreads and expected stock returns’,
Management Science 55, 1797–1812.
Bali, T. G., J. Hu & S. Murray (2015), Option implied volatility, skewness, and kurtosis
and the cross-section of expected stock returns. Working Paper.
27
Bali, T. G. & N. Cakici (2008), ‘Idiosyncratic volatility and the cross-section of expected
returns’, Journal of Financial and Quantitative Analysis 43, 29–58.
Bates, D. S. (1996), ‘Jumps and stochastic volatility: Exchange rate processes implicit in
deutsche mark options’, Review of Financial Studies 9, 69–107.
Black, F. & M. Scholes (1973), ‘The pricing of options and corporate liabilities’, Journal
of Political Economy 81, 637–654.
Bollerslev, T., G. Tauchen & H. Zhou (2009), ‘Expected stock returns and variance risk
premia’, Review of Financial Studies 22, 4463–4492.
Campbell, J., J. Hilscher & J. Szilagyi (2008), ‘In search of distress risk’, Journal of
Finance 63, 2899–2939.
Carhart, M. M. (1997), ‘On persistence in mutual fund performance’, Journal of Finance
52, 57–82.
Carr, P. & L. Wu (2009), ‘Variance risk premiums’, Review of Financial Studies 22, 1311–
1341.
Chakravarty, Sugato, Huseyin Gulen & Stewart Mayhew (2004), ‘Informed trading in
stock and option markets’, Journal of Finance 59(3), 1235–1257.
Chang, B. Y., P. Christoffersen & K. Jacobs (2013), ‘Market skewness risk and the cross-
section of stock returns’, Journal of Financial Economics 107, 46–68.
Chowdhry, B. & V. Nanda (1991), ‘Multimarket trading and market liquidity’, Review of
Financial Studies 4, 483–511.
Chung, Y. P., H. Johnson & M. J. Schill (2006), ‘Asset pricing when returns are non-
normal: Fama-french factors vs. higher-order systematic co-moments’, Journal of
Business 79, 923–940.
Cox, J. C. & M. Rubinstein (1985), Options Markets, NJ: Prentice-Hall.
Cox, J. C., S. A. Ross & M. Rubinstein (1979), ‘Option pricing: A simplified approach’,
Journal of Financial Economics 7, 229–263.
Cremers, M. & D. Weinbaum (2010), ‘Deviations from put-call parity and stock return
predictability’, Journal of Financial and Quantitative Analysis 45, 335–367.
28
Dennis, P., S. Mayhew & C. Stivers (2006), ‘Stock returns, implied volatility innovations,
and the asymmetric volatility phenomenon’, Journal of Financial and Quantitative
Analysis 41, 381–406.
Diamond, D. W. & R. E. Verrecchia (1987), ‘Constraints on short-selling and asset price
adjustment to private information’, Journal of Financial Economics 18, 277–311.
Diether, K. B., C. J. Malloy & A. Scherbina (2002), ‘Differences of opinion and the cross
section of stock returns’, Journal of Finance 57, 2113–2141.
Dittmar, R. F. (2002), ‘Nonlinear pricing kernels, kurtosis preference, and evidence from
the cross section of equity returns’, Journal of Finance 57, 369–403.
Doan, P., C.-T. Lin & R. Zurbruegg (2010), ‘Pricing assets with higher moments: evidence
from the australian and us stock markets’, Journal of International Financial
Markets Institutions and Money 20, 51–67.
Drechsler, I. & A. Yaron (2011), ‘Whats vol got to do with it’, Review of Financial Studies
24, 1–45.
Duan, J.-C. & J. Wei (2009), ‘Systematic risk and the price structure of individual equity
options’, Review of Financial Studies 22, 1981–2006.
Duffie, D., J. Pan & K. Singleton (2000), ‘Transform analysis and asset pricing for affine
jump-diffusions’, Econometrica 68, 1343–1376.
Easley, D., M. O’Hara & P. J. Srinivas (1998), ‘Option volume and stock prices: Evidence
on where informed traders trade’, Journal of Finance 53, 431–465.
Fama, E. F. & J. D. MacBeth (1973), ‘Risk, return, and equilibrium: Empirical tests’,
Journal of Political Economy 81, 607–636.
Fama, E. F. & K. R. French (1992), ‘The cross section of expected stock returns’, Journal
of Finance 47, 427–465.
Fama, E. F. & K. R. French (1993), ‘Common risk factors in the returns on stocks and
bonds’, Journal of Financial Economics 33, 3–56.
Fu, F. (2009), ‘Idiosyncratic risk and the cross-section of expected stock returns’, Journal
of Financial Economics 91, 24–37.
Gibbons, M., S. Ross & J. Shanken (1989), ‘A test of efficiency of a given portfolio’,
Econometrica 57, 1121–1152.
29
Guidolin, M. & A. Timmermann (2008), ‘International asset allocation under regime
switching, skew, and kurtosis references’, Review of Financial Studies 21, 889–935.
Harvey, C. R. & A. Siddique (2000), ‘Conditional skewness in asset pricing tests’, Journal
of Finance 55, 1263–1295.
Hasbrouck, J. (2009), ‘Trading costs and returns for u.s. equities: Estimating effective
costs from daily data’, Journal of Finance 64, 1445–1477.
Heston, S. L. (1993), ‘Closed-form solution for options with stochastic volatility with
applications to bonds and currency options’, Review of Financial Studies 6, 327–
343.
Jegadeesh, N. (1990), ‘Evidence of predictable behavior of security returns’, Journal of
Finance 45, 881–898.
Jegadeesh, N. & S. Titman (1993), ‘Returns to buying winners and selling losers:
Implications for stock market efficiency’, Journal of Finance 48, 65–91.
Jin, W., J. Livnat & Y. Zhang (2012), ‘Option prices leading equity prices: Do option
traders have an information advantage?’, Journal of Accounting Research 50, 401–
432.
Kimball, M. S. (1990), ‘Precautionary saving in the small and in the large’, Econometrica
58, 53–73.
Kraus, A. & R. H. Litzenberger (1976), ‘Skewness preference and the valuation of risk
assets’, Journal of Finance 31, 1085–1100.
Lehmann, B. N. (1990), ‘Fads, martingales, and market efficiency’, Journal of Finance
105, 1–28.
Lim, K.-G. (1989), ‘A new test of the three-moment capital asset pricing model’, Journal
of Financial and Quantitative Analysis 24, 205–216.
Loughran, T. & J. Ritter (1996), ‘Long-term market overreaction: The effect of low-priced
stocks’, Journal of Finance 51, 1959–1970.
Manaster, S. & R. J. Rendleman (1982), ‘Option prices as predictors of equilibrium stock
prices’, Journal of Finance 37, 1043–1058.
Nagel, S. (2005), ‘Short sales, institutional investors and the cross-section of stock returns’,
Journal of Financial Economics 78, 277–309.
30
Newey, Whitney K. & Kenneth D. West (1987), ‘A simple, positive semi-definite,
heteroskedasticity and autocorrelation consistent covariance matrix’, Econometrica
55(3), 703–708.
Ni, S.X., J. Pan & A.M. Poteshman (2008), ‘Volatility information trading in the option
market’, Journal of Finance 63, 1059–1091.
Scott, R. C. & P. A. Horvath (1980), ‘On the direction of preference for moments of higher
order than the variance’, Journal of Finance 35, 915–919.
Smith, D. R. (2007), ‘Conditional co-skewness and asset pricing’, Journal of Empirical
Finance 19, 91–119.
Tanaka, K., T. Yamada & T. Watanabe (2010), ‘Application of gram-charlier expansion
and bond moments for pricing of interest rates and credit risk’, Quantitative Finance
10, 645–662.
Xing, Y., X. Zhang & R. Zhao (2010), ‘What does the individual option volatility smirk
tell us about future equity returns?’, Journal of Financial and Quantitative Analysis
45, 641–662.
Yan, S. (2011), ‘Jump risk, stock returns, and slope of implied volatility smile’, Journal
of Financial Economics 99, 216–233.
Zhang, J. E. & Y. Xiang (2008), ‘The implied volatility smirk’, Quantitative Finance
8(3), 263–284.
31
A. Asset Pricing Implications of the Risk-netural
Higher-order Moments
For a more in-depth exploration of the relationship between option pricing and the option-
implied volatility curve, we first investigate the stochastic volatility (SV) model proposed
by Heston (1993). Specifically, we assume that the risk-neutral dynamics of the stock
price follows a system of stochastic differential equations given by
dSt = (r − q)Stdt+√vtStdW
(1)t , (16)
dvt = κ(θ − vt)dt+ σv√vtdW
(2)t , (17)
where E[dW
(1)t dW
(2)t
]= ρdt. Here, St denotes the stock price at time t, r is the
annualized risk-free rate under the continuous compounding rule, q is the annualized
continuous dividend yield, vt is the time-varying variance process whose evolution
follows the square-root process with a long-run variance of θ, a speed of mean reversion
κ, and a volatility of the variance process σv. In addition, W(1)t and W
(2)t are two
independent Brownian motions under the risk-neutral measure, and ρ represents the
instantaneous correlation between the two Brownian motions.
Based on our numerical experiments with base parameter set taken from Heston
(1993), Figure 5 demonstrates that IV slope reflects the leverage effect measured by the
correlation coefficient (ρ), while IV convexity represents the degree of a large contribution
of extreme events to the variance, i.e., tail risk, driven by the volatility of variance risk
(σv). Put simply, IV convexity contains the information about the volatility of stochastic
volatility (σv) and can be interpreted as a simple measure of the perceived kurtosis that
addresses the option-implied tail risk in the distribution of underlying stock returns.19
We next consider the impact of jumps in the dynamics of the underlying asset price.
For example, Bakshi, Cao & Chen (1997) show that jump components are necessary to
explain the observed shapes of implied volatility curves in practice. In the presence of
jump risk, the option-implied risk-neutral distribution of a stock return is a function of the
average jump size and jump volatility. To illustrate the implications of jump components
on the shape of the implied volatility curve, we consider the following stochastic-volatility
19On another note, IV convexity can be viewed as a component of variance risk premium (VRP),documented by Bakshi et al. (2003), Carr & Wu (2009), Bollerslev, Tauchen & Zhou (2009), Drechsler& Yaron (2011). According to Carr & Wu (2009), VRP consists of two components: (i) the correlationbetween the variance and the stock return and (ii) the volatility of the variance.
32
Figure 5: Impact of ρ and σv on IV slope and IV convexity in the SV Model
This figure shows the effect of ρ and σv on IV slope and IV convexity implied by
the SV model given by (16)-(17). The base parameter set (S, v0, κ, θ, σv, ρ, T, r, q) =
(100, 0.01, 2.0, 0.01, 0.1, 0.0, 0.5, 0.0, 0.0) is taken from Heston (1993). We take 0.8, 1.0 and 1.2
as the moneyness (K/S) points for IV slope and IV convexity.
−0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0−0.01
0
0.01
0.02
0.03
0.04
0.05
0.06
ρ
IV slopeIV Convexity
0 0.2 0.4 0.6 0.8 10
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
σV
IV SlopeIV Convexity
jump-diffusion (SVJ) model under the risk-neutral pricing measure given by20
dSt = (r − q − λµJ)Stdt+√vtStdW
(1)t + JStdNt, (18)
dvt = κ(θ − vt)dt+ σv√vtdW
(2)t , (19)
where E[dW
(1)t dW
(2)t
]= ρdt, Nt is an independent Poisson process with intensity λ > 0,
and J is the relative jump size with log(1 + J) ∼ N (log(1 + µJ)− 0.5σ2J , σ
2J). The SVJ
model can be taken as an extension of the SV model with the addition of log-normal
jumps in the underlying asset price dynamics.
As we can see from Figure 6 with base parameter set taken from Duffie et al. (2000),
our numerical analysis illustrates that IV slope is mainly driven by the average jump size
(µJ), whereas the jump size volatility (σJ) contributes mainly to IV convexity. From
this perspective, Yan (2011) argues that the implied-volatility spread between ATM call
and put options contain information about the perceived jump risk by investigating the
relationship between the implied-volatility spread and the cross-section of stock returns.
Strictly speaking, in the SVJ model framework, implied-volatility spread measure of Yan
(2011) simply captures the effect of µJ but ignores the information from σJ .
20The SVJ model given by (18)-(19) can be interpreted as a variation of the Bates (1996) model.
33
Figure 6: Impact of µJ and σJ on IV slope and IV convexity in the SVJ Model
This figure shows the effect of µJ and σJ on IV slope and IV convexity implied by the
SVJ model given by (18)-(19). The base parameter set (S, v0, κ, θ, σv, ρ, λ, µJ , σJ , T, r, q) =
(100, 0.0942, 3.99, 0.014, 0.27,−0.79, 0.11,−0.12, 0.15, 0.5, 0.0319, 0.0) is taken from Duffie, Pan
& Singleton (2000). We take 0.8, 1.0 and 1.2 as the moneyness (K/S) points for IV slope and
IV convexity.
−1 −0.8 −0.6 −0.4 −0.2 00
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
µJ
IV SlopeIV Convexity
0 0.2 0.4 0.6 0.8 10
0.05
0.1
0.15
0.2
0.25
σJ
IV SlopeIV Convexity
B. Firm Characteristics and Asset Pricing Factors
We define a firm’s size (Size) as the natural logarithm of the market capitalization (prc
× shrout ×1000), which is computed at the end of each month using CRSP data. When
computing book-to-market ratio (BTM), we match the yearly Book value of Equity or
BE [book value of common equity (CEQ) plus deferred taxes and investment tax credit
(txditc)] for all fiscal years ending in June at year t to returns starting in July of year
t-1, and dividing this BE by the market capitalization at month t-1. Hence, the book-to-
market ratio is computed on a monthly basis. Market betas (β) are estimated with rolling
regressions using the previous 36 monthly returns available up to month t-1 (a minimum
of 12 months) given by
(Ri −Rf )k = αi + βi(MKT −Rf )k + εi,k, (20)
where t− 36 ≤ k ≤ t− 1 on a monthly basis. Following Jegadeesh & Titman (1993), we
compute momentum (MOM) using cumulative returns over the past six months skipping
one month between the portfolio formation period and the computation period to exclude
the reversal effect. Momentum is also rebalanced every month and assumed to be held for
the next one month. Short-term reversal (REV) is estimated based on the past one-month
return as in Jegadeesh (1990) and Lehmann (1990). Motivated by Amihud (2002) and
Hasbrouck (2009), we define illiquidity (ILLIQ) as the average of the absolute value of
34
the stock return divided by the trading volume of the stock in thousand USD using the
past one-year’s daily data up to month t.
Following Harvey & Siddique (2000), we regress daily excess returns of individual
stocks on the daily market excess return and the daily squared market excess return using
a moving-window approach with an one-year window size. Specifically, we re-estimate the
regression model at each month-end, where the regression specification is given by
(Ri −Rf )k = αi + β1,i(MKT −Rf )k + β2,i(MKT −Rf )2k + εi,k, (21)
where t − 365 ≤ k ≤ t − 1 on a daily basis. Subsequently, the co-skewness (Coskew) is
defined as the coefficient of the squared market excess return. We require at least 225
trading days in a year to reduce the impact of infrequent trading.
We compute idiosyncratic volatility of stock returns according to Ang et al. (2006).
The daily excess returns of individual stocks over the last 30 days are regressed on daily
Fama-French three factors and momentum factors at the end of each month, where the
regression specification is given by
(Ri −Rf )k = αi + β1i(MKT −Rf )k + β2iSMBk + β3iHMLk + β4iWMLk + εk, (22)
where t − 30 ≤ k ≤ t − 1 on a daily basis. Idiosyncratic volatility is computed as the
standard deviation of the regression residuals. To reduce the impact of infrequent trading
on idiosyncratic volatility estimates, a minimum of 15 trading days in a month for which
CRSP reports both a daily return and non-zero trading volume is required.
We estimate systematic volatility using the method suggested by Duan & Wei (2009)
as v2sys = β2v2
M/v2 at the end of each month. Motivated by Dennis, Mayhew & Stivers
(2006), we compute idiosyncratic implied variance as v2idio = v2 − β2v2
M on a monthly
basis, where vM is the implied volatility of the S&P500 index option.
35
Tab
le1:
Des
crip
tive
Sta
tist
ics:
Opti
on-i
mplied
Vol
atilit
ies
Pan
elA
rep
orts
the
sum
mar
yst
atis
tics
ofth
efi
tted
imp
lied
vola
tili
ties
an
dfi
xed
del
tas
of
the
ind
ivid
ual
equ
ity
op
tion
sw
ith
on
em
onth
(30
day
s)to
exp
irat
ion
atth
een
dof
mon
thob
tain
edfr
om
Op
tion
Met
rics
.D
Sm
easu
res
the
deg
ree
of
acc
ura
cyin
the
fitt
ing
pro
cess
at
each
poin
t
and
com
pu
ted
by
the
wei
ghte
dav
erag
est
and
ard
dev
iati
on
s.P
an
elB
rep
ort
sth
esa
mp
led
istr
ibu
tion
of
un
iqu
efi
rms
by
year
an
din
du
stry
.E
ach
firm
isas
sign
edin
toon
eof
the
Fam
a-F
ren
ch12
ind
ust
ries
(FF
1-1
2)
class
ifica
tion
base
don
its
SIC
cod
e.T
he
sam
ple
per
iod
cove
rsJan
2000
toD
ec2013.
Pan
elA
.S
um
mar
yst
ati
stic
sof
the
fitt
edim
pli
edvo
lati
liti
esC
all
Op
tion
s∆
2025
3035
40
45
50
55
60
65
70
75
80
Mea
n0.
4942
0.48
170.
4739
0.4
696
0.4
677
0.4
676
0.4
694
0.4
730
0.4
779
0.4
841
0.4
917
0.5
019
0.5
157
std
ev0.
2774
0.27
640.
2755
0.2
746
0.2
735
0.2
724
0.2
722
0.2
732
0.2
744
0.2
763
0.2
783
0.2
808
0.2
840
DS
0.05
020.
0397
0.03
060.0
244
0.0
208
0.0
190
0.0
182
0.0
181
0.0
190
0.0
213
0.0
258
0.0
334
0.0
436
Pu
tO
pti
on
s∆
-80
-75
-70
-65
-60
-55
-50
-45
-40
-35
-30
-25
-20
Mea
n0.
4958
0.48
550.
4788
0.4
755
0.4
745
0.4
755
0.4
784
0.4
831
0.4
893
0.4
970
0.5
067
0.5
199
0.5
381
std
ev0.
2976
0.29
190.
2876
0.2
846
0.2
821
0.2
803
0.2
796
0.2
797
0.2
803
0.2
813
0.2
823
0.2
836
0.2
844
DS
0.04
520.
0369
0.02
890.0
231
0.0
197
0.0
181
0.0
178
0.0
183
0.0
198
0.0
228
0.0
285
0.0
382
0.0
514
Pan
elB
.S
am
ple
dis
trib
uti
on
by
year
and
ind
ust
ryY
ear
FF
1F
F2
FF
3F
F4
FF
5F
F6
FF
7F
F8
FF
9F
F10
FF
11
FF
12
Tota
l20
0012
159
214
94
56
700
165
81
263
229
332
586
2900
2001
8649
167
90
50
641
145
71
219
233
290
518
2559
2002
9145
170
88
58
539
94
73
235
246
333
493
2465
2003
9047
165
82
55
476
77
73
235
233
349
479
2361
2004
8949
178
100
53
515
94
79
248
271
372
522
2570
2005
9350
200
122
56
521
95
82
258
281
466
558
2782
2006
102
5423
1138
68
529
97
87
279
298
597
620
3100
2007
104
6324
8156
74
545
105
96
287
303
734
704
3419
2008
113
6324
6157
75
517
101
105
276
278
823
703
3457
2009
110
5623
2154
70
460
87
110
283
269
794
705
3330
2010
120
5623
7168
70
479
88
113
285
282
823
795
3516
2011
126
5825
4190
76
521
90
115
294
293
903
913
3833
2012
123
5625
1198
73
493
84
116
297
292
934
978
3895
2013
131
6325
2203
75
493
91
115
295
287
1069
1085
4159
36
Table 2: Descriptive Statistics: Shape of Implied Volatility Curves and FirmCharacteristics
This table reports the descriptive statistics of other variables used in the analysis. Panel A reports the descriptive statistics
of option-implied volatility, slope and convexity of implied volatility curves with one month (30 days) to expiration at the
end of each month. IVput(∆) and IVcall(∆) refer to the fitted put and call option-implied volatilities with one month
(30 days) to expiration, where ∆ is the options’ delta. IV level is defined as IV level = 0.5 × [IVput(−0.5) + IVcall(0.5)].
IV slope is computed by IV slope = IVput(−0.2) − IVput(−0.8). We adopt the definitions of IV spread and IV smirk
proposed by Yan (2011) and Xing et al. (2010), respectively. The convexity of option-implied volatility curve is calculated
by IV convexity = IVput(−0.2) + IVput(−0.8) − 2 × IVcall(0.5). Using daily options implied convexity of equity options
and S&P500 index option, we conduct time-series regressions per month to decompose option-implied convexity into
the systematic and idiosyncratic components given by IV convexityi,k = αi + βi × IV convexityS&P500,k + εi,k, where
t − 30 ≤ k ≤ t − 1 on a daily basis. The fitted values and residual terms are the systematic components (Convexitysys)
of options implied convexity and the idiosyncratic components (Convexityidio) of options implied convexity, respectively.
We select the observations at the end of each month. Panel B shows the descriptive statistics for each firm characteristics
(Size, BTM, Beta, MOM, REV, ILLIQ, Coskew). Size (ln mv) is computed at the end of each month and we define size
as natural logarithm of the market capitalization. When computing book-to-market ratio(BTM), we match the yearly BE
(book value of common equity (CEQ) plus deferred taxes and investment tax credit (txditc)) for all fiscal years ending
at year t-1 to returns starting in July of year t and this BE is divided by market capitalization at month t-1. Beta () is
estimated from time-series regressions of raw stock excess returns on the Rm-Rf by month-by-month rolling over past
three year (36 months) returns (a minimum of 12 months). Momentum (MOM) is computed based on past cumulative
returns over the past six months skipping one month between the portfolio formation period and the computation period to
exclude the reversal effect following Jegadeesh and Titman (1993). Reversal (REV) is computed based on past one-month
return following Jegadeesh (1990) and Lehmann (1990). Illiquidity (ILLIQ) is the average of the absolute value of stock
return divided by the trading volume of the stock in thousand USD calculated using the past one-year’s daily data up to
month t following Amihud (2002) and Hasbrouck(2009). Following Harvey and Siddique (2000), daily excess returns of
individual stocks are regressed on the daily market excess return and the daily squared market excess return using the
last one year data month by month given by (Ri − Rf )k = αi + β1,i(MKT − Rf )k + β2,i(MKT − Rf )2k + εi,k, where
t− 365 ≤ k ≤ t− 1 on a daily basis. The co-skewness of a stock is the coefficient of the squared market excess return. To
reduce the impact of infrequent trading on co-skewness estimates, we require a minimum of 255 trading days per month.
Panel A. The shape of option-implied volatility curveImplied Volatility Slope of Implied Volatility Curve Convexity of Implied Volatiltiy Curve
IV level IV slope IV spread IV smirk IV convexity IV convexitysys IV convexity idio
Mean 0.4739 0.0423 0.0090 0.0687 0.0952 0.0933 0.0019Stdev 0.2689 0.1614 0.1235 0.1413 0.2774 0.1703 0.2235
Median 0.4088 0.0422 0.0045 0.0516 0.0590 0.0655 -0.0053Observation 471112 471112 471112 471112 471112 471112 471112
Panel B. Variables of firm characteristicsSize BTM beta (β) MOM REV ILLIQ (×106) Coskew
Mean 19.4607 0.9186 1.1720 0.0966 0.0187 0.7808 -1.3516Median 19.3757 0.5472 0.9808 0.0461 0.0080 0.0183 -0.5745Stdev 2.1054 2.3613 1.1860 0.4515 0.1701 5.2832 15.2483
37
Table 3: Average returns sorted by option-implied volatility variables
This table reports the descriptive statistics of the average quintile portfolio returns (per month) sorted
by alternative measures of IV convexity. At the end of each month, all firms are assigned to one of five
portfolio groups based on alternative options implied convexity assuming that stocks are held for the next
one-month-period. This process is repeated in every month. Monthly stock returns are obtained from
the Center for Research in Security Prices (CRSP) with stocks traded on the NYSE (exchcd=1), Amex
(exchcd=2), and NASDAQ (exchcd=3). We use only common shares (shrcd in 10, 11). Stocks with a
price less than three dollars are excluded from the sample. “Q1-Q5” denotes an arbitrage portfolio that
buys a low option-implied convexity portfolio (Q1) and sells a high option-implied convexity portfolio
(Q5). The sample period covers Jan 2000 to Dec 2013. Value-weighted portfolio returns are weighted by
the lag of market capitalization of the underlying stocks. Numbers in parentheses indicate t-statistics.
Panel A. Equal-weighted and Value-weighted portfolio returns sorted by IV convexity
Quintile Avg # of firms Mean Stdev EW Return VW Return
Q1 (Low IV convexity) 432 -0.1364 0.2822 0.0208 0.0136Q2 423 0.0156 0.0317 0.0124 0.0061Q3 415 0.0637 0.0394 0.0098 0.0051Q4 404 0.1333 0.0672 0.009 0.0025
Q5 (High IV convexity) 389 0.398 0.3539 0.0074 0.0023
Q1-Q50.0134 0.0113(7.87) (5.08)
Panel B. Equal-weighted and Value-weighted portfolio returns sorted by IV spread
Quintile Avg # of firms Mean Stdev EW Return VW Return
Q1 (Low IV spread) 425 -0.0834 0.1491 0.0145 0.0109Q2 434 -0.0094 0.0131 0.0102 0.0067Q3 425 0.004 0.0109 0.0081 0.0044Q4 396 0.0189 0.0173 0.0066 0.004
Q5 (High IV spread) 384 0.1096 0.1844 0.0013 -0.0004
Q1-Q50.0131 0.0113(7.31) (3.91)
Panel C. Equal-weighted and Value-weighted portfolio returns sorted by IV smirk
Quintile Avg # of firms Mean Stdev EW Return VW Return
Q1 (Low IV smirk) 418 -0.0519 0.1462 0.0136 0.0112Q2 422 0.0292 0.0176 0.0099 0.0066Q3 417 0.0527 0.0202 0.0073 0.0035Q4 410 0.0845 0.0286 0.0066 0.0033
Q5 (High IV smirk) 396 0.2182 0.1863 0.0034 0.0019
Q1-Q50.0102 0.0094(5.05) (3.64)
38
Tab
le4:
Alt
ernat
ive
Mea
sure
sof
IVco
nve
xity
Th
ista
ble
rep
ort
sth
ed
escr
ipti
ve
stati
stic
sof
the
aver
age
qu
inti
lep
ort
folio
retu
rns
(per
month
)so
rted
by
alt
ern
ati
ve
mea
sure
sofIV
convexity
.IVput(∆
)an
dIVcall
(∆)
refe
rto
the
fitt
edp
ut
an
dca
llop
tion
-im
plied
vola
tiliti
esw
ith
on
em
onth
(30
days)
toex
pir
ati
on
,w
her
e∆
isth
eop
tion
s’del
ta.
At
the
end
of
each
month
,all
firm
sare
ass
ign
edto
on
e
of
five
port
folio
gro
up
sb
ase
don
alt
ern
ati
ve
op
tion
sim
plied
convex
ity
ass
um
ing
that
stock
sare
hel
dfo
rth
en
ext
on
e-m
onth
-per
iod
.T
his
pro
cess
isre
pea
ted
inev
ery
month
.
Month
lyst
ock
retu
rns
are
ob
tain
edfr
om
the
Cen
ter
for
Res
earc
hin
Sec
uri
tyP
rice
s(C
RS
P)
wit
hst
ock
str
ad
edon
the
NY
SE
(exch
cd=
1),
Am
ex(e
xch
cd=
2),
an
dN
AS
DA
Q
(exch
cd=
3).
We
use
on
lyco
mm
on
share
s(s
hrc
din
10,
11).
Sto
cks
wit
ha
pri
cele
ssth
an
thre
ed
ollars
are
excl
ud
edfr
om
the
sam
ple
.“Q
1-Q
5”
den
ote
san
arb
itra
ge
port
folio
that
bu
ys
alo
wop
tion
-im
plied
convex
ity
port
folio
(Q1)
an
dse
lls
ah
igh
op
tion
-im
plied
convex
ity
port
folio
(Q5).
Th
esa
mp
lep
erio
dco
ver
sJan
2000
toD
ec2013.
Pan
elA
rep
ort
s
des
crip
tive
stati
stic
sof
the
equal-
wei
ghte
dan
dvalu
e-w
eighte
daver
age
port
folio
retu
rns
(per
month
)so
rted
by
alt
ern
ati
veIV
convexity
mea
sure
sby
vary
ing
ITM
an
dO
TM
poin
ts,
defi
ned
as
p∆
1-c
50-p
∆2=IVput
(−∆
1)+IVput
(−∆
1)-
2×IVcall
(0.5
),w
her
e0.2
≤∆
1≤
0.4
5an
d0.5
5≤
∆2≤
0.8
.V
alu
e-w
eighte
dp
ort
folio
retu
rns
are
wei
ghte
dby
the
lag
of
mark
etca
pit
aliza
tion
of
the
un
der
lyin
gst
ock
s.In
Pan
elB
,th
ed
efin
itio
ns
ofIVconvexity(I
I)-IVconvexity(V
I)
are
giv
enby
(9)-
(13).
Nu
mb
ers
inp
are
nth
eses
ind
icate
t-st
ati
stic
s.
Pan
elA
.A
lter
nat
ive
mea
sure
sof
IVconvexity
by
vary
ing
ITM
and
OT
Mp
oints
p25
-c50
-p75
p30
-c50
-p70
p35
-c50
-p65
p40-c
50-p
60
p45-c
50-p
55
Quin
tile
Mea
nE
WR
etV
WR
etM
ean
EW
Ret
VW
Ret
Mea
nE
WR
etV
WR
etM
ean
EW
Ret
VW
Ret
Mea
nE
WR
etV
WR
etQ
1(L
ow)
-0.1
304
0.02
020.
0136
-0.1
364
0.0
199
0.01
06-0
.138
50.
0199
0.01
09-0
.1388
0.0
198
0.0
106
-0.1
395
0.0
204
0.0
116
Q2
0.00
100.
0120
0.00
58-0
.007
90.0
123
0.00
64-0
.013
00.
0124
0.00
61-0
.0155
0.0
124
0.0
063
-0.0
163
0.0
122
0.0
069
Q3
0.04
210.
0095
0.00
520.
0292
0.0
096
0.00
490.
0206
0.00
950.
0056
0.01
470.0
096
0.0
057
0.0
110
0.0
090
0.0
052
Q4
0.09
770.
0093
0.00
230.
0755
0.0
089
0.00
310.
0604
0.00
850.
0034
0.04
980.0
090
0.0
032
0.0
428
0.0
088
0.0
034
Q5
(Hig
h)
0.31
200.
0080
0.00
290.
2644
0.0
084
0.00
350.
2313
0.00
880.
0019
0.20
920.0
083
0.0
008
0.1
967
0.0
086
0.0
012
Q1-
Q5
0.01
220.
0106
0.0
115
0.00
710.
0111
0.00
900.0
115
0.0
098
0.0
118
0.0
104
(7.7
4)(4
.85)
(7.3
2)(3
.36)
(7.1
1)(3
.89)
(7.0
3)
(4.0
7)
(6.7
9)
(3.8
7)
Pan
elB
.A
lter
nat
ive
mea
sure
sof
IVconvexity
bet
wee
nca
lls
and
puts
IVconvexity(I
I)
IVconvexity(I
II)
IVconvexity(I
V)
IVconvexity(V
)IVconvexity(V
I)
Quin
tile
Mea
nStd
evA
vg
Ret
Mea
nStd
evA
vg
Ret
Mea
nStd
evA
vg
Ret
Mea
nStd
evA
vg
Ret
Mea
nStd
evA
vg
Ret
Q1
(Low
)-0
.056
00.
1615
0.00
95-0
.056
20.1
588
0.01
54-0
.040
70.
1191
0.01
90-0
.0571
0.1
627
0.0
177
-0.0
714
0.2
098
0.0
203
Q2
0.01
620.
0131
0.00
920.
0139
0.0
138
0.01
250.
0227
0.01
340.
0110
0.01
660.0
130
0.0
140
0.0
240
0.0
158
0.0
104
Q3
0.04
710.
0214
0.00
820.
0434
0.0
229
0.00
930.
0501
0.02
000.
0098
0.04
700.0
213
0.0
095
0.0
570
0.0
217
0.0
095
Q4
0.09
570.
0414
0.00
810.
0904
0.0
416
0.00
970.
0923
0.03
500.
0096
0.09
550.0
413
0.0
099
0.1
139
0.0
439
0.0
090
Q5
(Hig
h)
0.27
500.
2040
0.00
570.
2533
0.1
893
0.01
280.
2350
0.15
660.
0103
0.27
470.2
034
0.0
086
0.3
376
0.2
663
0.0
103
Q1-
Q5
0.00
380.0
026
0.008
70.0
091
0.0
100
(2.4
9)(1
.76)
(5.1
0)
(6.0
9)
(5.1
4)
39
Tab
le5:
Quin
tile
por
tfol
ios
sort
edby
(firm
-siz
e,b
ook
-to-
mar
ket
rati
o,m
arke
tβ
)an
dIV
con
vexi
ty
Th
ista
ble
rep
ort
sth
eaver
age
month
lyre
turn
sof
twen
tyfi
ve
dou
ble
-sort
edp
ort
folios,
firs
tso
rted
by
firm
chara
cter
isti
cvari
ab
les
(firm
size
,b
ook-t
o-m
ark
etra
tio,
mark
et
β)
an
dIV
convexity
on
am
onth
lyb
asi
s,an
dth
ensu
b-s
ort
edacc
ord
ing
toIV
convexity
wit
hin
each
qu
inti
lep
ort
folio
into
on
eof
the
five
port
foli
os
on
am
onth
lyb
asi
s.
All
firm
-sp
ecifi
cvari
ab
les
are
as
des
crib
edin
Tab
le2.
Sto
cks
are
ass
um
edto
be
hel
dfo
ron
em
onth
,an
dp
ort
folio
retu
rns
are
equ
ally-w
eighte
d.
Month
lyst
ock
retu
rns
are
ob
tain
edfr
om
the
Cen
ter
for
Res
earc
hin
Sec
uri
tyP
rice
s(C
RS
P)
wit
hst
ock
str
ad
edon
the
NY
SE
(exch
cd=
1),
Am
ex(e
xch
cd=
2)
an
dN
AS
DA
Q(e
xch
cd=
3).
We
use
on
lyco
mm
on
share
s(s
hrc
din
10,
11).
Sto
cks
wit
ha
pri
cele
ssth
an
thre
ed
ollars
are
excl
ud
edfr
om
the
sam
ple
.“Q
1-Q
5”
den
ote
san
arb
itra
ge
port
folio
that
bu
ys
alo
w
IVconvexity
port
folio
an
dse
lls
ah
ighIV
convexity
port
folio
inea
chch
ara
cter
isti
cp
ort
folio.
Th
esa
mp
leco
ver
sJan
2000
toD
ec2013.
Nu
mb
ers
inp
are
nth
eses
ind
icate
st-
stati
stic
s.
Aver
age
Ret
urn
IVconvexity
Siz
eQ
uin
tile
sB
TM
Qu
inti
les
Mark
etβ
Qu
inti
les
S1
(Sm
all)
S2
S3
S4
S5
(Larg
e)B
1(L
ow
)B
2B
3B
4B
5(H
igh
)β
1(L
ow
)β
2β
3β
4β
5(H
igh
)Q
1(L
ow
)0.0
255
0.0
164
0.0
112
0.0
122
0.0
093
0.0
106
0.0
148
0.0
171
0.0
225
0.0
337
0.0
157
0.0
189
0.0
189
0.0
213
0.0
263
Q2
0.0
205
0.0
118
0.0
111
0.0
093
0.0
064
0.0
035
0.0
089
0.0
121
0.0
153
0.0
274
0.0
100
0.0
118
0.0
125
0.0
132
0.0
148
Q3
0.0
160
0.0
111
0.0
080
0.0
109
0.0
071
0.0
063
0.0
080
0.0
106
0.0
134
0.0
195
0.0
096
0.0
121
0.0
099
0.0
109
0.0
102
Q4
0.0
139
0.0
097
0.0
081
0.0
081
0.0
057
0.0
033
0.0
060
0.0
082
0.0
126
0.0
189
0.0
084
0.0
109
0.0
109
0.0
098
0.0
084
Q5
(Hig
h)
0.0
068
0.0
034
0.0
064
0.0
065
0.0
034
0.0
009
0.0
028
0.0
063
0.0
091
0.0
160
0.0
076
0.0
083
0.0
072
0.0
089
0.0
050
Q1-Q
50.0
187
0.0
130
0.0
048
0.0
057
0.0
059
0.0
097
0.0
121
0.0
108
0.0
134
0.0
177
0.0
081
0.0
106
0.0
118
0.0
124
0.0
213
(6.4
9)
(5.8
6)
(2.2
6)
(3.2
3)
(3.9
7)
(4.8
2)
(5.7
6)
(5.4
2)
(5.6
7)
(5.1
9)
(4.7
3)
(6.1
5)
(6.1
3)
(5.1
3)
(6.3
2)
Aver
age
nu
mb
erof
firm
s
IVconvexity
Siz
eQ
uin
tile
sB
TM
Qu
inti
les
Bet
aQ
uin
tile
sS
1(S
mall)
S2
S3
S4
S5
(Larg
e)B
1(L
ow
)B
2B
3B
4B
5(H
igh
)β
1(L
ow
)β
2β
3β
4β
5(H
igh
)Q
1(L
ow
)82
82
82
82
82
75
75
75
75
75
76
77
76
75
70
Q2
83
83
83
83
83
76
76
76
76
76
79
79
79
78
76
Q3
83
83
83
83
83
76
76
76
76
76
79
79
79
78
76
Q4
83
83
83
83
83
76
76
76
76
76
78
79
79
78
76
Q5
(Hig
h)
82
82
82
82
82
75
75
75
75
75
76
77
77
76
73
40
Table 6: Quintile portfolios sorted by (MOM, REV, ILLIQ, Coskew) and IV convexity
This table reports the average monthly returns of a double-sorted quintile portfolio formed based on (momentum, reversal,
illiquidity, co-skewness) and IV convexity. All firm-specific variables are as described in Table 2. The sample covers Jan
2000 to Dec 2013 with stocks traded on the NYSE (exchcd=1), Amex (exchcd=2) and NASDAQ (exchcd=3). For each
month, stocks are sorted into five groups based on (momentum, reversal, liquidity, co-skewness) and then subsorted within
each quintile portfolio into one of the five portfolios according to IV convexity. Stocks are assumed to be held for one
month, and portfolio returns are equally-weighted. We use only common shares (shrcd in 10, 11). Stocks with a price less
than three dollars are excluded from the sample. “Q1-Q5” denotes an arbitrage portfolio that buys a low IV convexity
portfolio and sells a high IV convexity portfolio in each of momentum, reversal, illiquidity, and coskew portfolios. Numbers
in parentheses indicate t-statistics.
Average Return
IV convexityMOM Quintiles REV Quintiles
M1 (Loser) M2 M3 M4 M5 (Winner) R1 (Low) R2 R3 R4 R5 (High)Q1 (Low) 0.0225 0.0157 0.0130 0.0147 0.0273 0.0237 0.0170 0.0160 0.0136 0.0134
Q2 0.0107 0.0094 0.0103 0.0100 0.0185 0.0160 0.0132 0.0096 0.0092 0.0083Q3 0.0025 0.0060 0.0078 0.0087 0.0183 0.0123 0.0119 0.0098 0.0075 0.0065Q4 -0.0012 0.0071 0.0070 0.0086 0.0177 0.0084 0.0092 0.0087 0.0066 0.0068
Q5 (High) -0.0029 0.0036 0.0072 0.0082 0.0166 0.0061 0.0095 0.0070 0.0055 0.0013
Q1-Q50.0254 0.0121 0.0058 0.0066 0.0108 0.0176 0.0076 0.0089 0.0081 0.0122(8.36) (5.92) (3.76) (4.51) (5.39) (6.99) (3.64) (4.69) (4.49) (5.73)
IV convexityILLIQ Quintiles Coskew Quintiles
I1 (Low) I2 I3 I4 I5 (High) C1 (Low) C2 C3 C4 C5 (High)Q1 (Low) 0.0108 0.0157 0.0165 0.0222 0.0302 0.0128 0.0169 0.0131 0.0132 0.0140
Q2 0.0081 0.0104 0.0135 0.0158 0.0231 0.0115 0.0111 0.0116 0.0085 0.0082Q3 0.0064 0.0100 0.0101 0.0127 0.0174 0.0098 0.0099 0.0087 0.0098 0.0057Q4 0.0053 0.0083 0.0110 0.0116 0.0162 0.0056 0.0086 0.0092 0.0074 0.0037
Q5 (High) 0.0051 0.0097 0.0072 0.0069 0.0109 -0.0003 0.0079 0.0091 0.0056 0.0013
Q1-Q50.0057 0.0060 0.0093 0.0153 0.0193 0.0130 0.0090 0.0040 0.0076 0.0127(2.88) (3.01) (4.28) (6.12) (6.39) (4.60) (3.83) (2.39) (4.18) (5.07)
Average number of firms
IV convexityMOM Quintiles REV Quintiles
M1 (Loser) M2 M3 M4 M5 (Winner) R1 (Low) R2 R3 R4 R5 (High)Q1 (Low) 80 80 80 80 79 77 79 79 79 78
Q2 80 80 80 80 80 80 80 80 80 80Q3 80 80 80 80 80 80 80 80 81 80Q4 80 80 80 80 80 79 80 80 80 80
Q5 (High) 80 80 80 79 78 78 79 79 79 78
IV convexityILLIQ Quintiles Coskew Quintiles
I1 (Low) I2 I3 I4 I5 (High) C1 (Low) C2 C3 C4 C5 (High)Q1 (Low) 83 81 83 83 78 73 75 75 75 74
Q2 87 81 82 83 77 76 77 77 77 76Q3 85 80 82 81 74 76 77 77 77 76Q4 82 80 80 80 73 76 77 77 77 76
Q5 (High) 74 74 75 75 74 74 76 76 76 75
41
Tab
le7:
Quin
tile
por
tfol
ios
sort
edby
(IV
slop
e,IV
spre
ad,
IVsm
irk
)an
dIV
con
vexi
ty
Th
ista
ble
rep
ort
sth
eaver
age
month
lyre
turn
sof
ad
ou
ble
-sort
edqu
inti
lep
ort
folio
form
edb
ase
don
(IV
slope
,IV
spread,IV
smirk
)an
dIV
convexity
.P
ort
folios
are
sort
edin
five
gro
up
sat
the
end
of
each
month
base
don
(IV
slope
,IV
spread,IV
smirk
)fi
rst
an
dth
ensu
b-s
ort
edin
tofi
ve
gro
up
sb
ase
donIV
convexity
.O
pti
on
svola
tility
slop
esare
com
pu
ted
byIVslope
=IVput(−
0.2
)−IVput(−
0.8
),IVspread
=IVput(−
0.5
)−IVcall
(0.5
),an
dIV
smirk
isco
mp
ute
db
ase
don
the
met
hod
sugges
ted
by
Xin
get
al.
(2010),
resp
ecti
vel
y.S
tock
sare
hel
dfo
ron
em
onth
,an
dp
ort
folio
retu
rns
are
equ
al-
wei
ghte
d.
Month
lyst
ock
retu
rns
are
ob
tain
edfr
om
the
Cen
ter
for
Res
earc
hin
Sec
uri
tyP
rice
s
(CR
SP
)w
ith
stock
str
ad
edon
the
NY
SE
(exch
cd=
1),
Am
ex(e
xch
cd=
2)
an
dN
AS
DA
Q(e
xc
hcd
=3).
We
use
on
lyco
mm
on
share
s(s
hrc
din
10,
11).
Sto
cks
wit
ha
pri
cele
ss
than
thre
ed
ollars
are
excl
ud
edfr
om
the
sam
ple
.“Q
1-Q
5”
den
ote
san
arb
itra
ge
port
folio
that
bu
ys
alo
wIV
convexity
port
folio
an
dse
lls
ah
igh
IVconvexity
port
folio
in
each
(IV
slope
,IV
spread,IV
smirk
)p
ort
folio.
Th
esa
mp
lep
erio
dco
ver
sJan
2000
toD
ec2013.
Nu
mb
ers
inpare
nth
eses
ind
icate
t-st
ati
stic
s.
Aver
age
Ret
urn
IVconvexity
IVslope
Qu
inti
les
IVspread
Qu
inti
les
IVsm
irk
Qu
inti
les
S1(L
ow
)S
2S
3S
4S
5(H
igh
)S
1(L
ow
)S
2S
3S
4S
5(H
igh
)S
1(L
ow
)S
2S
3S
4S
5(H
igh
)Q
1(L
ow
)0.0
267
0.0
270
0.0
178
0.0
160
0.0
151
0.0
320
0.0
163
0.0
123
0.0
138
0.0
130
0.0
116
0.0
098
0.0
110
0.0
110
0.0
113
Q2
0.0
165
0.0
164
0.0
115
0.0
111
0.0
136
0.0
228
0.0
132
0.0
092
0.0
085
0.0
122
0.0
112
0.0
073
0.0
048
0.0
097
0.0
077
Q3
0.0
126
0.0
101
0.0
079
0.0
089
0.0
123
0.0
189
0.0
102
0.0
073
0.0
067
0.0
091
0.0
090
0.0
063
0.0
054
0.0
063
0.0
074
Q4
0.0
079
0.0
070
0.0
081
0.0
070
0.0
089
0.0
150
0.0
111
0.0
091
0.0
074
0.0
059
0.0
101
0.0
060
0.0
068
0.0
054
0.0
028
Q5
(Hig
h)
0.0
039
0.0
085
0.0
066
0.0
113
0.0
087
0.0
167
0.0
104
0.0
078
0.0
076
0.0
028
0.0
042
0.0
048
0.0
061
0.0
052
-0.0
004
Q1-Q
50.0
227
0.0
185
0.0
112
0.0
047
0.0
064
0.0
153
0.0
059
0.0
045
0.0
063
0.0
102
0.0
074
0.0
049
0.0
049
0.0
058
0.0
116
(6.8
4)
(6.3
0)
(4.5
9)
(2.1
4)
(2.5
2)
(5.2
1)
(3.2
2)
(2.6
1)
(3.1
7)
(3.7
7)
(3.2
7)
(2.1
8)
(2.2
1)
(1.9
3)
(3.9
7)
Aver
age
nu
mb
erof
firm
s
IVconvexity
IVsl
op
eQ
uin
tile
sIV
spre
ad
Qu
inti
les
IVsm
irk
Qu
inti
les
S1(L
ow
)S
2S
3S
4S
5(H
igh
)S
1(L
ow
)S
2S
3S
4S
5(H
igh
)S
1(L
ow
)S
2S
3S
4S
5(H
igh
)Q
1(L
ow
)82
85
86
88
86
85
88
85
82
80
46
47
48
49
48
Q2
80
82
86
87
86
87
85
83
77
75
47
47
47
47
46
Q3
79
83
86
86
84
85
87
85
78
76
46
47
46
46
43
Q4
77
81
83
84
80
86
89
87
79
76
45
45
44
44
42
Q5
(Hig
h)
77
79
80
80
77
82
86
85
79
77
42
43
41
41
38
42
Tab
le8:
Ave
rage
por
tfol
iore
turn
sso
rted
by
syst
emat
ican
did
iosy
ncr
atic
com
pon
ents
ofIV
con
vexi
ty
Pan
elA
rep
ort
the
des
crip
tive
stati
stic
sof
the
aver
age
port
foli
ore
turn
sso
rted
by
syst
emati
cco
mp
on
ents
(convexitysys)
an
did
iosy
ncr
ati
cco
mp
on
ents
(convexityid
io)
ofIV
convexity
.A
llfi
rm-s
pec
ific
vari
ab
les
are
as
des
crib
edin
Tab
le2.
At
the
end
of
each
month
,all
firm
sare
ass
ign
edin
toon
eof
five
port
folio
gro
up
sbase
don
IVconvexity
,
(convexitysys,convexityid
io)
an
dw
eass
um
est
ock
sare
hel
dfo
rth
en
ext
on
e-m
onth
-per
iod
.T
his
pro
cess
isre
pea
ted
inev
ery
month
.P
an
elB
rep
ort
sth
eaver
age
month
lyre
turn
s
of
ad
ou
ble
-sort
edqu
inti
lep
ort
folio
usi
ng
syst
emati
cco
mp
on
ents
ofIV
convexity
(convexitysys)
an
did
iosy
ncr
ati
cco
mp
on
ents
ofIV
convexity
(convexityid
io).
Port
folios
are
sort
edin
tofi
ve
gro
up
sat
the
end
of
each
month
base
don
(convexitysys,convexityid
io)
firs
t,an
dth
ensu
b-s
ort
edin
tofi
ve
gro
up
sb
ase
donIV
convexity
.S
tock
sare
hel
dfo
ron
e
month
,an
dp
ort
folio
retu
rns
are
equ
al-
wei
ghte
d.
Sto
cks
are
ass
um
edto
be
hel
dfo
ron
em
onth
,and
port
folio
retu
rns
are
equ
ally-w
eighte
d.
Month
lyst
ock
retu
rns
are
ob
tain
ed
from
the
Cen
ter
for
Res
earc
hin
Sec
uri
tyP
rice
s(C
RS
P)
wit
hst
ock
str
ad
edon
the
NY
SE
(exch
cd=
1),
Am
ex(e
xch
cd=
2)
an
dN
AS
DA
Q(e
xch
cd=
3).
We
use
on
lyco
mm
on
share
s(s
hrc
din
10,
11)
of
wh
ich
stock
pri
ceex
ceed
sth
ree
dollars
are
excl
ud
edfr
om
the
sam
ple
.“Q
1-Q
5”
den
ote
san
arb
itra
ge
port
folio
that
bu
ys
alo
wIV
convexity
port
folio
an
dse
lls
ah
ighIV
convexity
port
folio
inea
ch(convexitysys,convexityid
io)
port
folio.
Th
esa
mp
lep
erio
dco
ver
sJan
2000
toD
ec2013.
Nu
mb
ers
inp
are
nth
eses
ind
icate
t-st
ati
stic
s.
Pan
elA
.S
yst
emat
ican
did
iosy
ncr
atic
com
pon
ents
ofIV
convexity
and
aver
age
por
tfol
iore
turn
IVconvexity
convexitysys
convexityid
io
Avg
#of
firm
sM
ean
Std
evA
vg
Ret
Avg
#of
firm
sM
ean
Std
evA
vg
Ret
Q1
(Low
)423
-0.0
434
0.1
517
0.0
199
420
-0.2
113
0.2
402
0.0
174
Q2
426
0.0
373
0.0
199
0.0
111
419
-0.0
518
0.0
383
0.0
117
Q3
420
0.0
682
0.0
260
0.0
100
413
-0.0
039
0.0
330
0.0
116
Q4
410
0.1
145
0.0
425
0.0
094
408
0.0
470
0.0
483
0.0
101
Q5
(Hig
h)
385
0.2
773
0.2
113
0.0
089
403
0.2
379
0.2
781
0.0
087
Q1-Q
50.0
110
0.0
086
(6.5
6)
(5.7
2)
Pan
elB
.D
ou
ble
sort
edqu
inti
lep
ortf
olio
byconvexity sys
andconvexity idio
IVconvexity
Aver
age
Ret
urn
c sys1
c sys2
c sys3
c sys4
c sys5
c idio
1c i
dio
2c i
dio
3c i
dio
4c i
dio
5Q
1(L
ow
)0.0
319
0.0
158
0.0
156
0.0
143
0.0
155
0.0
152
0.0
145
0.0
130
0.0
111
0.0
049
Q2
0.0
224
0.0
122
0.0
091
0.0
091
0.0
095
0.0
129
0.0
107
0.0
105
0.0
081
0.0
044
Q3
0.0
188
0.0
090
0.0
079
0.0
101
0.0
094
0.0
115
0.0
093
0.0
079
0.0
075
0.0
059
Q4
0.0
131
0.0
102
0.0
088
0.0
077
0.0
070
0.0
101
0.0
064
0.0
095
0.0
068
0.0
007
Q5
(Hig
h)
0.0
140
0.0
086
0.0
090
0.0
059
0.0
029
0.0
049
0.0
065
0.0
079
0.0
057
-0.0
033
Q1-Q
50.0
179
0.0
072
0.0
066
0.0
084
0.0
126
0.0
102
0.0
080
0.0
051
0.0
053
0.0
082
(5.9
3)
(3.4
9)
(3.2
4)
(4.0
7)
(5.4
5)
(3.1
6)
(3.6
6)
(2.6
3)
(2.7
1)
(3.3
0)
IVconvexity
Aver
age
nu
mb
erof
firm
sc s
ys1
c sys2
c sys3
c sys4
c sys5
c idio
1c i
dio
2c i
dio
3c i
dio
4c i
dio
5Q
1(L
ow
)553
775
776
762
665
631
767
713
741
737
Q2
691
676
736
741
644
783
753
659
732
790
Q3
677
646
713
713
604
801
753
696
755
754
Q4
702
696
726
709
561
755
754
730
745
709
Q5
(Hig
h)
686
766
767
721
478
605
672
686
663
590
43
Tab
le9:
Shor
t-se
llin
gco
nst
rain
tsan
din
form
atio
nas
ym
met
ry
Th
ista
ble
rep
ort
sth
eaver
age
month
lyre
turn
sof
twen
tyfi
ve
dou
ble
-sort
edp
ort
folios,
firs
tso
rted
by
the
pre
vio
us
qu
art
erly
per
centa
ge
of
share
sou
tsta
nd
ing
hel
dby
inst
itu
tion
s
(fro
mth
eT
hom
son
Fin
an
cial
Inst
itu
tion
al
Hold
ings
(13F
)d
ata
base
),p
revio
us
qu
art
er’s
an
aly
stco
ver
age
(fro
mI/
B/E
/S
),p
revio
us
qu
art
er’s
an
aly
stfo
reca
std
isp
ersi
on
(fro
m
I/B
/E
/S
)an
dth
ensu
b-s
ort
edacc
ord
ing
toIV
convexity
wit
hin
each
qu
inti
lep
ort
folio
into
on
eof
the
five
port
folios
on
am
onth
lyb
asi
s.F
ollow
ing
Cam
pb
ell
etal.
(2008)
an
d
Nagel
(2005),
we
calc
ula
teth
esh
are
of
inst
itu
tion
al
ow
ner
ship
by
sum
min
gth
est
ock
hold
ings
of
all
rep
ort
ing
inst
itu
tion
sfo
rea
chst
ock
inea
chqu
art
er.
An
aly
stfo
reca
st
dis
per
sion
ism
easu
red
as
the
scale
dst
an
dard
dev
iati
on
of
I/B
/E
/S
an
aly
sts’
curr
ent
fisc
al
yea
rea
rnin
gs
per
share
fore
cast
sby
the
met
hod
inD
ieth
eret
al.
(2002).
Sto
cks
are
ass
um
edto
be
hel
dfo
ron
em
onth
,an
dp
ort
foli
ore
turn
sare
equ
ally-w
eighte
d.
Month
lyst
ock
retu
rns
are
ob
tain
edfr
om
the
Cen
ter
for
Res
earc
hin
Sec
uri
tyP
rice
s(C
RS
P)
wit
h
stock
str
ad
edon
the
NY
SE
(exch
cd=
1),
Am
ex(e
xch
cd=
2)
an
dN
AS
DA
Q(e
xch
cd=
3).
We
use
on
lyco
mm
on
share
s(s
hrc
din
10,
11)
of
wh
ich
pri
ceex
ceed
s.th
ree
dollars
.
Q1-Q
5d
enote
san
arb
itra
ge
port
folio
that
bu
ys
alo
wIV
convexity
port
folio
an
dse
lls
ah
ighIV
convexity
port
folio
inea
chch
ara
cter
isti
cp
ort
folio.
Th
esa
mp
leco
ver
sJan
2000
toD
ec2013.
Nu
mb
ers
inp
are
nth
eses
ind
icate
st-
stati
stic
s.
Aver
ave
Ret
urn
IVconvexity
F-1
3fi
lin
gs(
Inst
itu
tion
al
Hold
ings)
An
aly
stC
over
age
An
aly
stF
ore
cast
Dis
per
sion
F1(L
ow
)F
2F
3F
4F
5(H
igh
)A
C1(L
ow
)A
C2
AC
3A
C4
AC
5(H
igh
)A
D1(L
ow
)A
D2
AD
3A
D4
AD
5(H
igh
)Q
1(L
ow
)0.0
228
0.0
215
0.0
210
0.0
195
0.0
150
0.0
230
0.0
228
0.0
195
0.0
169
0.0
177
0.0
225
0.0
207
0.0
215
0.0
190
0.0
151
Q2
0.0
150
0.0
141
0.0
132
0.0
117
0.0
084
0.0
156
0.0
151
0.0
122
0.0
107
0.0
094
0.0
161
0.0
139
0.0
118
0.0
115
0.0
089
Q3
0.0
103
0.0
083
0.0
103
0.0
112
0.0
075
0.0
129
0.0
115
0.0
102
0.0
098
0.0
084
0.0
140
0.0
108
0.0
087
0.0
091
0.0
014
Q4
0.0
088
0.0
124
0.0
117
0.0
101
0.0
066
0.0
118
0.0
122
0.0
098
0.0
087
0.0
081
0.0
144
0.0
126
0.0
095
0.0
088
0.0
035
Q5
(Hig
h)
0.0
064
0.0
096
0.0
086
0.0
124
0.0
085
0.0
086
0.0
087
0.0
086
0.0
078
0.0
063
0.0
145
0.0
121
0.0
099
0.0
061
-0.0
011
Q1-Q
50.0
164
0.0
120
0.0
124
0.0
071
0.0
065
0.0
145
0.0
141
0.0
109
0.0
091
0.0
114
0.0
081
0.0
086
0.0
116
0.0
129
0.0
162
(5.1
8)
(4.4
4)
(5.7
8)
(3.3
8)
(3.3
4)
(4.9
3)
(5.0
2)
(4.3
6)
(4.6
0)
(5.2
1)
(3.7
6)
(4.4
6)
(4.8
7)
(5.3
4)
(5.4
6)
Aver
age
nu
mb
erof
firm
s
IVconvexity
F-1
3fi
lin
gs
(In
stit
uti
on
al
Hold
ings)
An
aly
stC
over
age
An
aly
stF
ore
cast
Dis
per
sion
F1(L
ow
)F
2F
3F
4F
5(H
igh
)A
C1(L
ow
)A
C2
AC
3A
C4
AC
5(H
igh
)A
D1(L
ow
)A
D2
AD
3A
D4
AD
5(H
igh
)Q
1(L
ow
)49
51
52
51
51
60
70
74
73
76
71
71
70
67
61
Q2
49
54
54
53
55
58
71
76
75
78
75
74
73
71
64
Q3
49
54
54
53
54
59
71
75
74
78
75
74
73
70
64
Q4
48
53
53
52
54
57
69
74
74
78
73
73
71
68
61
Q5
(Hig
h)
46
51
51
49
51
56
66
70
71
75
70
69
67
63
55
44
Tab
le10
:T
ime-
seri
esan
alysi
sof
CA
PM
and
Fam
a-F
rench
thre
ean
dfo
ur
fact
orm
odel
s
Th
ista
ble
pre
sents
the
coeffi
cien
tes
tim
ates
of
CA
PM
an
dF
am
a-F
ren
chth
ree
an
dfo
ur
fact
or
mod
els
for
month
lyex
cess
retu
rns
onIV
convexity
qu
inti
les
por
tfol
ios.
Fam
a-F
ren
chfa
ctor
s[R
M−R
f,
small
mark
etca
pit
ali
zati
on
min
us
big
(SM
B),
an
dh
igh
book-t
o-m
ark
etra
tio
min
us
low
(HM
L),
and
mom
entu
mfa
ctor
(UM
D)]
are
obta
ined
from
Ken
net
hF
ren
ch’s
web
site
.T
he
sam
ple
per
iod
cove
rsJan
2000
toD
ec2013
wit
hst
ock
str
ad
edon
the
NY
SE
(exch
cd=
1),
Am
ex(e
xch
cd=
2)an
dN
AS
DA
Q(e
xch
cd=
3).
Sto
cks
wit
ha
pri
cele
ssth
an
thre
ed
oll
ars
are
excl
ud
edfr
om
the
sam
ple
,an
d
New
ey&
Wes
t(1
987)
adju
sted
t-st
atis
tics
are
rep
ort
edin
squ
are
bra
cket
s.
Model
Facto
rsIV
convexity
convexitysys
convexityidio
Q1
Q2
Q3
Q4
Q5
Q1-Q
5Q1
Q2
Q3
Q4
Q5
Q1-Q
5Q1
Q2
Q3
Q4
Q5
Q1-Q
5
CAPM
Alp
ha
0.0146
0.0069
0.0042
0.0032
0.0014
0.0132
0.0138
0.0056
0.0043
0.0035
0.0031
0.0107
0.0113
0.0061
0.0061
0.0044
0.0026
0.0087
(5.60)
(4.75)
(2.88)
(2.04)
(0.66)
(7.72)
(5.51)
(3.96)
(2.83)
(2.05)
(1.53)
(6.39)
(4.70)
(3.92)
(4.41)
(2.88)
(1.20)
(5.62)
MKTRF
1.4472
1.2075
1.2505
1.3057
1.3830
0.0642
1.4207
1.2283
1.2811
1.3393
1.3338
0.0870
1.4086
1.2623
1.2307
1.2831
1.4110
-0.0024
(23.11)
(31.82)
(36.13)
(34.53)
(27.20)
(1.38)
(25.71)
(33.44)
(35.29)
(30.46)
(27.65)
(2.06)
(25.73)
(30.01)
(33.77)
(35.37)
(29.21)
(-0.06)
Adj.R
20.7920
0.8968
0.9065
0.9007
0.8500
0.0126
0.0126
0.9057
0.9003
0.8858
0.8494
0.0295
0.0295
0.8932
0.9128
0.9017
0.8442
¡0.0001
FF3
Alp
ha
0.0124
0.0048
0.0023
0.0011
-0.0010
0.0134
0.0118
0.0039
0.0024
0.0011
0.0004
0.0114
0.0087
0.0038
0.0042
0.0025
0.0005
0.0082
(5.87)
(4.52)
(2.34)
(1.03)
(-0.64)
(7.75)
(6.00)
(3.69)
(2.47)
(0.90)
(0.26)
(7.48)
(4.60)
(3.62)
(4.12)
(2.49)
(0.31)
(5.36)
MKTRF
1.3027
1.1149
1.1512
1.2026
1.2650
0.0377
1.2767
1.1361
1.1724
1.2278
1.2313
0.0454
1.2713
1.1624
1.1475
1.1776
1.2796
-0.0083
(23.58)
(37.48)
(44.49)
(40.90)
(28.50)
(0.85)
(28.30)
(38.33)
(46.30)
(34.29)
(29.01)
(1.38)
(25.92)
(36.13)
(38.38)
(43.02)
(29.43)
(-0.20)
SM
B0.6328
0.4493
0.4557
0.4824
0.5520
0.0808
0.6199
0.4238
0.4955
0.5324
0.5111
0.1088
0.6300
0.4786
0.4016
0.4772
0.5825
0.0475
(5.33)
(8.39)
(9.74)
(8.46)
(5.76)
(1.44)
(5.55)
(8.44)
(11.93)
(7.58)
(5.01)
(2.41)
(5.85)
(8.64)
(7.18)
(9.60)
(5.91)
(0.91)
HM
L0.0211
0.1346
0.0717
0.1006
0.1138
-0.0927
-0.0094
0.0685
0.0690
0.1377
0.1863
-0.1957
0.0995
0.1277
0.1146
0.0573
0.0386
0.0609
(0.23)
(2.71)
(1.70)
(2.44)
(1.67)
(-1.27)
(-0.12)
(1.42)
(1.73)
(2.51)
(2.70)
(-3.48)
(1.19)
(2.39)
(2.65)
(1.31)
(0.61)
(0.84)
Adj.R
20.8685
0.9541
0.9642
0.9585
0.9133
0.0491
0.0491
0.9572
0.9654
0.9508
0.9063
0.1781
0.1781
0.9527
0.9576
0.9625
0.9159
¡0.0001
FF4
Alp
ha
0.0132
0.0050
0.0024
0.0014
-0.0003
0.0136
0.0126
0.0040
0.0026
0.0015
0.0011
0.0115
0.0094
0.0042
0.0044
0.0028
0.0011
0.0083
(7.33)
(5.10)
(2.48)
(1.54)
(-0.29)
(7.61)
(7.55)
(3.94)
(2.90)
(1.48)
(0.90)
(7.32)
(5.85)
(4.63)
(4.58)
(2.93)
(0.88)
(5.32)
MKTRF
1.1125
1.0580
1.1264
1.1304
1.1112
0.0013
1.1050
1.1005
1.1200
1.1355
1.0804
0.0246
1.1044
1.0793
1.0974
1.1272
1.1380
-0.0336
(18.13)
(34.90)
(37.00)
(38.48)
(27.91)
(0.02)
(19.37)
(34.59)
(42.41)
(34.70)
(29.53)
(0.54)
(20.55)
(40.32)
(35.53)
(36.75)
(25.76)
(-0.69)
SM
B0.7588
0.4870
0.4722
0.5302
0.6539
0.1049
0.7336
0.4474
0.5302
0.5935
0.6110
0.1226
0.7406
0.5336
0.4348
0.5106
0.6763
0.0643
(7.28)
(11.09)
(9.97)
(11.01)
(8.59)
(1.51)
(7.50)
(9.27)
(13.78)
(10.08)
(7.54)
(2.09)
(7.64)
(12.92)
(8.60)
(11.41)
(8.01)
(1.04)
HM
L-0
.0143
0.1240
0.0671
0.0871
0.0852
-0.0995
-0.0413
0.0619
0.0592
0.1206
0.1582
-0.1995
0.0684
0.1122
0.1053
0.0480
0.0123
0.0561
(-0.22)
(3.02)
(1.64)
(2.52)
(1.86)
(-1.36)
(-0.68)
(1.35)
(1.78)
(3.06)
(3.89)
(-3.30)
(1.20)
(3.10)
(2.89)
(1.22)
(0.23)
(0.80)
UM
D-0
.3321
-0.0994
-0.0433
-0.1261
-0.2686
-0.0635
-0.2999
-0.0622
-0.0915
-0.1612
-0.2635
-0.0364
-0.2916
-0.1452
-0.0875
-0.0880
-0.2474
-0.0441
(-4.35)
(-4.37)
(-1.77)
(-7.22)
(-11.29)
(-0.97)
(-3.96)
(-2.40)
(-4.13)
(-7.41)
(-9.08)
(-0.61)
(-4.59)
(-5.23)
(-3.86)
(-4.63)
(-6.43)
(-0.84)
Adj.R
20.9232
0.9619
0.9654
0.9694
0.9553
0.0675
0.0675
0.9600
0.9713
0.9675
0.9497
0.1813
0.9324
0.9681
0.9635
0.9679
0.9499
0.0025
45
Table 11: Fama-MacBeth regressions
This table reports the averages of month-bymonth Fama & MacBeth (1973) cross-sectional regression coefficient estimates
for individual stock returns on IV convexity and control variables. The cross-section of expected stock returns is regressed
on control variables. Control variables include market β estimated by Fama & French (1992), size (ln mv), book-to-market
(btm), momentum (MOM), reversal (REV), illiquidity (ILLIQ), IV slope, idiosyncratic risk (idio risk), implied volatility
level (IV level), systematic volatility (v2sys), idiosyncratic implied variance (v2idio). All firm-specific variables are as
described in Table 2. Following Ang et al. (2006), we regress daily excess returns of individual stocks on the Fama-French
four factors daily in every month as (Ri − Rf )k = αi + β1i(MKT − Rf )k + β2iSMBk + β3iHMLk + β4iWMLk + εk,
where t − 30 ≤ k ≤ t − 1 on a daily basis. The idiosyncratic volatility of a stock (idio risk) is computed as the standard
deviation of the regression residuals. Systematic volatility is estimated by the method suggested by Duan & Wei (2009)
as v2sys = β2v2m/v2. Idiosyncratic implied variance is computed as v2idio = v2 − β2v2M , where vm is the implied volatility
of S&P500 index option, as suggested by Dennis et al. (2006). The daily factor data are downloaded from Kenneth R.
French’s website. To reduce the impact of infrequent trading on idiosyncratic volatility estimates, we require a minimum of
15 trading days in a month for which CRSP reports both a daily return and non-zero trading volume. The sample period
covers Jan 2000 to Dec 2013 with stocks traded on the NYSE (exchcd=1), Amex (exchcd=2) and NASDAQ (exchcd=3)
and stocks with a price less than three dollars are excluded from the sample. The adjusted t-statistics based on Newey
& West (1987) using lag of 3 are reported for the time-series average of coefficients. An intercept is included in each
specification but not reported in the table. Numbers in parentheses indicate t-statistics.
Variable MODEL 1 MODEL 2 MODEL 3 MODEL 4 MODEL 5 MODEL 6 MODEL 7 MODEL 8IV -0.015*** -0.017*** -0.015*** -0.016*** -0.015*** -0.017*** -0.015*** -0.017***
convexity (-6.60) (-6.19) (-6.70) (-6.38) (-6.61) (-6.17) (-6.59) (-6.17)IV -0.004 -0.004 -0.005 -0.005slope (-0.98) (-1.26) (-1.41) (-1.30)
Beta 0.001 0.001 0.000 0.000 0.000 0.000 0.001 0.001(0.33) (0.34) (0.20) (0.22) (0.25) (0.27) (0.41) (0.45)
Log(MV)-0.002** -0.002** -0.002*** -0.002*** -0.002*** -0.002*** -0.002*** -0.002***(-2.39) (-2.30) (-2.86) (-2.75) (-3.26) (-3.11) (-2.85) (-2.71)
Btm 0.004** 0.004** 0.004** 0.004** 0.003** 0.003** 0.004** 0.004**(2.14) (2.15) (2.24) (2.27) (2.15) (2.18) (2.18) (2.22)
MOM 0.001 0.001 0.001 0.001 0.001 0.001(0.15) (0.18) (0.28) (0.31) (0.21) (0.24)
REV-0.017*** -0.017*** -0.017*** -0.017*** -0.017*** -0.017***
(-3.47) (-3.45) (-3.48) (-3.46) (-3.43) (-3.42)
ILLIQ -0.007 -0.007 0.002 0.001 0.001 0.000(-0.37) (-0.36) (0.09) (0.07) (0.04) (0.02)
idio risk -0.110** -0.104*(-2.00) (-1.90)
IV -0.007 -0.007level (-1.05) (-1.01)
v2sys-0.002 -0.002(-1.52) (-1.58)
v2idio-0.008* -0.008*(-1.82) (-1.81)
Adj.R2 0.050 0.051 0.064 0.065 0.068 0.068 0.067 0.068
46
Tab
le12
:Sub-p
erio
dro
bust
nes
san
alysi
s
Th
ista
ble
rep
orts
the
aver
age
equ
al-w
eigh
ted
retu
rns
of
the
qu
inti
lep
ort
foli
os
form
edon
IVconvexity
for
vari
ou
ssu
bsa
mp
les.
We
chec
kth
e
rob
ust
nes
sof
our
resu
lts
by
div
idin
gth
een
tire
sam
ple
per
iod
into
the
exp
an
sion
an
dco
ntr
act
ion
sub
-per
iod
s(i
)by
takin
gth
em
edia
nva
lue
of
the
Ch
icag
oF
edN
atio
nal
Act
ivit
yIn
dex
(CF
NA
I)as
ath
resh
old
an
d(i
i)by
usi
ng
the
NB
ER
Bu
sin
ess
cycl
ed
um
my
vari
ab
le,
wh
ich
takes
on
efo
rth
e
contr
acti
onp
erio
dan
d0
oth
erw
ise.
At
the
end
of
each
month
,w
eso
rtst
ock
sbase
donIV
convexity
into
five
qu
inti
les
an
dfo
rman
equ
al-
wei
ghte
d
por
tfol
ioth
atb
uys
alo
wIV
convexity
por
tfol
ioan
dse
lls
ah
ighIV
convexity
port
foli
o.
We
ass
um
est
ock
sare
hel
dfo
rth
en
ext
on
e-m
onth
-per
iod
.
Th
isp
roce
ssis
rep
eate
dfo
rev
ery
mon
th.
The
sam
ple
cove
rsJan
2000
toD
ec2013.
Ch
icag
oF
edN
atio
nal
Act
ivit
yIn
dex
(CF
NA
I)N
BE
RB
usi
nes
sC
ycl
eD
um
my
Exp
ansi
onp
erio
dC
ontr
act
ion
per
iod
sE
xp
an
sion
per
iod
sC
ontr
act
ion
per
iods
(84
mon
ths)
(84
month
s)(1
42
month
s)(2
6m
onth
s)
Qu
inti
leM
ean
Std
evA
vg
Mea
nS
tdev
Avg
Mea
nS
tdev
Avg
Mea
nS
tdev
Avg
Ret
Ret
Ret
Ret
Q1
(Low
)-0
.125
60.
2701
0.0127
-0.1
478
0.2
939
0.0
288
-0.1
394
0.2
897
0.0
142
-0.1
584
0.2
685
0.0
103
Q2
0.01
440.
0252
0.0
101
0.0
167
0.0
371
0.0
146
0.0
150
0.0
288
0.0
119
0.0
181
0.0
459
0.0
016
Q3
0.05
930.
0270
0.0
085
0.0
680
0.0
483
0.0
111
0.0
618
0.0
342
0.0
100
0.0
756
0.0
610
-0.0
019
Q4
0.12
600.
0473
0.0
071
0.1
406
0.0
815
0.0
108
0.1
301
0.0
576
0.0
091
0.1
537
0.1
044
-0.0
063
Q5
(Hig
h)
0.37
380.
3319
0.0
053
0.4
221
0.3
729
0.0
094
0.3
915
0.3
441
0.0
046
0.4
863
0.4
582
-0.0
133
Q1-
Q5
0.0074
0.0
194
0.0
096
0.0
236
(4.9
2)
(6.6
5)
(5.8
8)
(3.8
7)
47